Model Releases
1964 signals across 219 briefings.
2026-10-01
- π΄Clef: Open Weights decision model by Cloudflare
- π΄Putting One Kimi Among Four Claudes, Can the Claudes Identify Kimi?
- π΄Qwen4Exp: add MTP by am17an Β· Pull Request #29761 Β· ggml-org/llama.cpp
- π΄Oct 1, 2026 Announcements Barclays scales Claude to upgrade operations and improve client experience
- π΄How Albertsons Companies is reimagining retail from the inside out
- π‘GPT-Synopsys: Frontier Intelligence to Revolutionize Chip Design
- π‘Jeff-Qwen3.5-0.8B v1.2 + 9 LoRA adapters: put it in front of Qwen3.8-27B for 38Γ faster decisions and +8.7 points accuracy, for under 2 GB extra memory
- π‘As a Plus subscriber, Sol 6.1 is a game changer
- π‘PewDiePie has just launched a new uncensored AI model (based on Qwen 3.5 9B)
- π‘@AnthropicAI: In physics, an βimpedance mismatchβ occurs when two systems each work well but are poorly matched. In this Science Blog guest post, Harvard physicist Matthew Schwartz argues that something similar is
- π’Update #2: Post training yandex/AliceAI-80B-A3B [instruct!] from scratch
- π’Cheeky cheeky Google, I see what you're doing, it was flash all along, Deepmind researcher working on scaling Gemini, appears to be deleted now.
- π’SupersonicLabs/Julia-1 (2,556 downloads)
2026-09-30
- π΄Gemini 4 from straight from the horses mouth
- π΄Introducing Gemini 4 Argon
- π΄add GLM-5.3-Flash (GLM5-Next) support by timkhronos Β· Pull Request #27773 Β· ggml-org/llama.cpp
- π΄ChatGPT vs Perplexity vs Gemini on the same 120 tool questions: 42% had zero overlap.
- π΄This is a hot mess
- π‘DeepSeek now trained on Ascend 950
- π‘Gemini 4 Argon (High): Intelligence, Performance and Price Analysis
- π‘add GLM-5.3-Flash (GLM5-Next) support (#27773) Β· ggml-org/llama.cpp@649dcb1
- π‘BBC: OpenAI agents get rebrand - as 'dots' - while safety worries delay new model
- π‘Responsible Release of AI-Generated Mathematics
- π’is the AI product flood actually a bubble, or just the messy part of a wave that ends up mattering?
- π’Qwen3.8-27B-pi: Effort-Ordered Reasoning for Agentic Coding
- π’New models used to come out every 10 weeks. Now it's every 11 days.
- π’Gemini 4 Argon Releases
- π’You can now run Qwen 3.8 27B on AMD NPUs via FastFlowLM (at a killer 1 tps decode)
2026-09-29
- π΄OpenAI launches GPT-6.1 Sol
- π΄A company ran 8 identical AI societies for weeks with different models and just published what happened. Some of it is genuinely unsettling.
- π΄Looks like the era of subsidised compute is coming to an end. The old ChatGPT Pro $200 20x plan will be halved. The new $500 plan will have similar limits as the (old) $200 plan.
- π΄GPT-6.1 Sol - Apparently near-Astra performance for complex work at a lower cost.
- π΄DevDay 2026 Recap
- π‘Deepseek Harness app is out now!!!
- π‘@AnthropicAI: Claude Sonnet 5.5 is now available:
- π‘[Release] GSQ-RCO GGUFs for Qwen3.8-Flash-Next, plus a 50% expert-pruned Coder build at ~1.89 bpw
- π‘@AnthropicAI: What do you want from AI? Weβre launching a new study with Anthropic Interviewer to learn more about your experiences using AI, what role you want it to play in your life and the world, and what you w
- π‘RAM Offloading with vLLM - tcclaviger appreciation post
- π’Why Does This 2022 Anime Feel Like an AI Prophecy?
- π’Dots isnβt avaliable in Europe
- π’Show HN: TurboGPT: train 22KiB transformer in 13s
- π’Recommended replacements for glm 4.7 flash
- π’Limited compute, targeting CVPR: rerun experiments for statistically strong numbers or focus on writing? [D]
2026-09-28
- π΄Claude Sonnet 5.5 Released
- π΄GPT-3 is discontinued today
- π΄Nvidia launches new tool to keep AI agents from going rogue
- π΄Guys I promise It wasn't me ππ
- π΄Claude Code told me a task was done and everything worked. It was not true
- π‘You can just merge Qwen3.6 & 3.8 27B to get decent performance with much fewer token generation
- π‘Why are Chinese labs so focused on open models?
- π‘Free, open-source AI engineering course where you build each algorithm by hand: 523 lessons, now as EPUB/PDF books [P]
- π‘What model sits between Qwen 3.8 27b and Flash next for coding?
- π‘@gdb: Pinned: mega upgrade for GPT Voice, which can now use tools and is available in Work:
- π’XingChen-AGI/TeleOCR (27,904 downloads)
- π’Hey folks, Opus 5.5 from Claude simply destroys any currently existing GPT model
- π’Qwen 3.8 is a workhorse
- π’OpenAI Scraps Release of New AI Model Over Safety Concerns (WSJ Exclusive)
- π’I emptied my trash with a Blender project inside β ChatGPT located the SSD sectors and rebuilt it
2026-09-27
- π΄The Opus 5.5 posts about motion graphics are cool but Qwen 27B made this on a 4090
- π΄Are there machine learning subfields that are becoming irrelevant (or is irrelevant)? [D]
- π΄42x Faster Prompt Lookup Drafting in llama.cpp
- π΄Introducing Vitruvian, a genetic optimization model for optimizing human embryo DNA and reportedly being able to increase the IQ of a person by 14 points, nearly a standard deviation which is incredibly significant
- π΄Adding logit penalty for "wait", "maybe" and "perhaps" to Qwen models improves their accuracy
- π‘where do people put human review in an agent workflow?
- π‘How do you guys give your models web browsing capabilities?
- π‘Don't trust frontier models when asking about budget hardware!
- π’... so, yeah.
- π’AI labs need business-style controls on testing and release, and the recent incidents show why
- π’Annoyed of ChatGPT App changing their UI every couple of days
- π’Swift 1.5 Qwen3.8 27b (A must-have for low thinking!)
2026-09-26
- π΄Swift 1.5 27b: Swift Qwen just got faster
- π΄My friend gave Claude Code and Codex agents a way to talk to each other. Once this went over a hundred agents they reinvented bureaucracy.
- π΄Introducing KoboldCpp Agent (and a plea for help)
- π΄Claude Opus 5.5, GPT-6 Sol, GPT-6 Luna, And A New Price War (6 Minute Read)
- π΄Palo Alto Networks Unveils AI-Powered Cybersecurity Service Using Claude, GPT Models (3 Minute Read)
- π‘DeepSeek Elastic Compute (DSec)
- π‘New to playing around with local ai. why are they free?
- π‘If your AI agents could collaborate directly with each other across devices & harnesses, what would you have them work on together?
- π‘GPT 6 (non Astra) is performing much worse than expected?
- π‘Sonnet 5.5, Which Already Supposedly Beats GPT-6 Sol, Has Had a Last-Minute Upgrade With Release Expected Monday
- π’Improved and fixed template for GPT-OSS (again). Includes preserve_thinking and fix for Unsloth-induced bug
- π’The new agent, O, releasing on DevDay is rumored to not be included in Plus plans.
- π’Run Qwen3.8+Flash-Next and tiny models on Apple Silicon up to 3x faster
- π’A great example of how inaccurate AI can still be
2026-09-25
- π΄Qwengram-0.8B: I transferred Qwen3.8 Flash-Nextβs n-gram memory into Qwen3.5-0.8B β 5.05% lower validation perplexity
- π΄Swift1.5-Qwen3.8-Flash-Next is phenomenal vs. base 3.8-Flash!
- π΄Proaction boosts sales 60% and saves 75+ hours with Codex
- π΄Compass is coming to the cloud Gain access to our state-of-the-art search and retrieval engine, without worrying about the infrastructure Sep 25, 2026 5 min read
- π΄I went to Legoland yesterday with the kids - very impressive to see all the cities theyβve built with lego. Whoβs job is that?!
- π‘gpt is down for everyone right?
- π‘Ollaya β Ollama for open-source, Jev-style decision models
- π‘4-5 days replacing Claude w Qwen 3.8 Next
- π‘@AnthropicAI: New on the Science Blog: Yes, Claude can do Nine Loops. Theoretical physicists predict how particles behave using formulas called scattering amplitudes. These are notoriously hard to compute, so resea
- π‘Introducing: GPT-6 Rock
- π’Claude Fable 5.1 completed a frontier nine-loop particle-physics calculation experts had worked toward for years, with researchers mostly just telling it βkeep goingβ while it built, debugged and ran the entire workflow
- π’Edge0/Audio8-ASR-Infinite (2,853 downloads)
- π’XiaomiMiMo/MiMo-V2.6-Pro-RL (42,062 downloads)
2026-09-24
- π΄My Fiance is Convinced AI will likely cause a Catastrophic or Extinction-Type Event in the Next Few Years - How Justified Are His Fears?
- π΄JEV almost dead: CLM vs JEV
- π΄GPT-6 Astra conquered KSPβs rocket simulator by landing on every surface world, and even mined fuel on Moho to make the trip home
- π΄ChatGPT Ads expands to Southeast Asia and Taiwan
- π΄Introducing Gemini 3.8 Live with Live Avatar
- π‘I trained an AI on 25 years of my own writing and told it not to be helpful. Here's what happened.
- π‘GPT-6 Sol is NOT the replacement of GPT-5.6 Sol...
- π‘@reach_vb: Massive QoL update: ChatGPT Voice can now use plugins AND run on GPT-6 Astra, Sol & Luna, directly in ChatGPT Work on web and mobile. Your email, calendar, Slack + docs, decks, sites and spreadsheets,
- π‘Is Claude really so much better than ChatGPT?
- π‘Agents to agents with a computer, from OpenClaw to Grok Bot & Muse
- π’TaichuAI/ZDTaichu5.0-9B (8,313 downloads)
- π’Finally got AI coding agents running directly on my phone (no termux/root) kinda impressed ngl
- π’ChatGPT Helped Tumbler Ridge Shooter Focus on Guns, Tactics, and Terror
- π’vulkan: int8 coopmat1 matmul implementation for AMD RDNA3 and RDNA4 (β¦ Β· ggml-org/llama.cpp@70c4e15
- π’Qwen3.8 27b practical modeling for 3d printing
2026-09-23
- π΄Claude discovered a novel enzyme system with properties reminiscent of CRISPR
- π΄Gemini 3.8 text-to-speech
- π΄Claude Code reads AGENTS.md only when telemetry is on [fixed]
- π΄Harvey turns legal context into stronger drafts with GPT-6 Astra
- π΄How invideo improves color grading 3x with GPTβ6 Astra
- π‘@OpenAI: Weβre demonstrating how frontier models have continued to improve in realistic mental health conversations with MentalHealthBench. This new open benchmark was built with input from more than 80 mental
- π‘i think someone got Trump to rename artificial intelligence as super intelligence, in order to make it harder to talk about Artificial Super Intelligence
- π‘The AI you test in the afternoon may not be the AI you test at night, even with the same name
- π‘Anthropic CEO Dario Amodei: AI Biology Will Go From Weak To Superhuman Within A Few Years
- π‘@AnthropicAI: In the Democratic Republic of the Congo, global health organizations including @CEPIvaccines, @WHOAFRO, and @inrb_kinshasa are using Claude to accelerate their response to an outbreak of an unusual Eb
- π’AgentFlayer (enterprise agents, Zenity Labs)
- π’BFL releases FLUX 3 Action: a 7B robot model
- π’AlexWortega/openjev (0 downloads)
- π’Using uncensored models makes working less of a headache
2026-09-22
- π΄Introducing GPT-6 Sol and Luna
- π΄Qwen 4 Announced at Apsara Conference
- π΄Introducing Claude Opus 5.5, 40% Cheaper and Smarter Than Ever Before
- π΄How did the Chinese labs catch up so fast if most of the early training work was in the US?
- π΄Ngram and world knowledge - why are we just building a coding model?
- π‘Saw this today about how GPT Astra is the reason a professional is giving up on their career as a Three.js expert in 3d modelling.
- π‘My local llm when I tell it to do any changes to my vLLM service
- π‘Understanding and Enhancing Kimi Delta Attention [R]
- π‘Did Alibaba abandon 35B A3B?
- π‘Claude Opus 5.5 Intelligence, Performance and Price Analysis (Max)
- π’Comfy-Org/Qwen-Image-2.1 (1,429,925 downloads)
- π’GPT6 Astra is picking up business spend fast
- π’Launch HN: Coverage Cat (YC S22) β Umbrella insurance via your personal agent
- π’Now Opus 5.5 is 58 on Artificial Analysis , how long do you have wait until an open model hits 58?
- π’GPT-6 Astra just made a leap in musical reasoning: it wrote a complex four-part Bach-style piece with no harmony-rule errors and techniques no previous AI managed
2026-09-21
- π΄Introducing Grok 4.7
- π΄Clarification on the Qwen-image-2.1 license
- π΄Higgsfield AI ships new video features in a day with GPT-6 Astra
- π΄How V7 gives AI agents institutional memory
- π΄Quite a few new model releases expected this week, latest rumors attached
- π‘[Leaks] GPT-6 Is Already Legacy Software. SHIP 6.1. π
- π‘[MASSIVE RELEASE] Supra2-IMG - a tiny 100M text-to-image model - SOTA quality and open release!
- π‘@Alibaba_Qwen: Thanks @sgl_project for the day-0 support! π SGLang-Diffusion now serves Qwen-Image-2.1: text-to-image generation, multi-image editing, and transparent RGBA output. Try it out! π¨
- π‘@Alibaba_Qwen: Qwen-Image-2.1 Γ @HuggingApps: live demo on Spaces! πΌ One single checkpoint for generation and editing. Try it in your browser, no setup needed. π
- π‘Deepseek training 2T and plans 8T model
- π’My ChatGPT voice mode just mocked my gf π
- π’yandex/AliceAI-Foundation-80B-A3B-Base: Russian-developed competitor to Qwen 35B and DeepSeek V4 Flash
- π’abenzerps/Qwen-Image-2.1-GGUF (33,232 downloads)
- π’XiaomiMiMo/MiMo-V2.6-Distill-Qwen-9B
2026-09-20
- π΄Qwen-Image-2.1 released!
- π΄What is JEV and what is it used for?
- π΄Duckdb Skills For Claude Code (4 Minute Read)
- π΄Introducing Home Mcp: Enabling Your Agent To Interact With Your Home (2 Minute Read)
- π΄Gemini In Google Workspace Can Now Connect Directly To Salesforce, Hubspot, Asana, And More (4 Minute Read)
- π‘ChatGPT has now surpassed Instagram by site visits
- π‘Would you buy a Qwen3.8-27B Taalas chip for $1k if it could run at 7,000 TPS?
- π‘This coming week could have a couple of model releases
- π‘CUDA: enable sparse fa for qwen4 by am17an Β· Pull Request #28770 Β· ggml-org/llama.cpp
- π‘Anyone else get auto-downgraded off the 20Γ plan with most of your quota unused?
- π’Qwen/Qwen-Image-2.1 (183 downloads)
- π’is astra 6 supposed to do this?
2026-09-19
- π΄Introducing the Australian Youth Safety Blueprint
- π΄We coveredJevβs launch on Wednesday, and they have completely taken over the timeline, with 36M views of their launch video (by comparison,OpenAIβs Navier Stokesresult got74Mviews, andAnthropicβs Fabl
- π΄Bespoke Nimble: LoRA finetune of Qwen3.5-9B, using contrastive data curation. (close but sllightly lower on benchmarks)
- π΄SemIf (fka OpenJev)(HF): 4B and 35B causal Qwen3.5 backbone with a tiny three-class NLI classifier on the last token.comparison vs Laya
- π΄Kev-0.5B: LoRA adapter + a small readout head on top of Qwen2.5-0.5B.
- π‘@GoogleDeepMind: Weβre introducing Gemini 3.8 Live and 3.8 Live Extended Thinking β our best conversational AI. The models talk, think, and handle tasks in the background without breaking your flow. π§΅
- π‘What happens to our money if banking system gets hacked by AI and data gets wiped out?
- π‘Thank you Google ππ₯³π₯³
- π‘Built a home server from an old PC with GPU upgrade. Qwen3.8 27B runs at ~30 tokens per second.
- π‘ChatGPT-6 Astra cracks 108-year-old unsolved WWI German code for the first time
- π’convaiinnovations/laya (0 downloads)
- π’I asked GPT-6 Astra to optimize itself so it stops hitting usage limits
- π’Made an alternative to gpt researcher,need your validation,cuts 80percent token limit...
- π’harshatheg/Qwen-2.5-1B-RLCD (0 downloads)
- π’Qwen-3.8-Flash-Next on 1x RTX 5090: TG=50 t/s, PP=2300 t/s - with FreeToken
2026-09-18
- π΄Claude Code now reads AGENTS.md if there is no Claude.md
- π΄ZCode was allegedly caught uploading workspace/.git records to the cloud.
- π΄Sep 18, 2026 Announcements Partnering with Accenture on embedded evaluation
- π΄How Cooley is accelerating IPO work with ChatGPT
- π΄Hex turns complex analysis into visual reports with GPTβ6 Astra
- π‘Astra ported the original Portal game to my iPhone 13 Pro
- π‘US government website used AI search tool (Qwen) from China that FBI said copied Anthropic
- π‘Show HN: Cactus Needle 3: 8-29MB automation models can match DeepSeek V4 Flash
- π‘Built a job change monitoring agent for our sales team, simpler than i expected
- π‘Question: UkisAI Swift Ternary Bonsai 2 27B?
- π’Claude now leads 26% of AI development at Anthropic
- π’What studies isolate back-and-forth LLM interaction from one-way sharing and self-refinement [D]
- π’NGL, Iβm hyped to see if Qwen3.8 27b can make me a sandwich. Instant buy for me.
2026-09-17
- π΄I literally built the Jev architecture one year back and completely open-sourced it with model, dataset and paper
- π΄Thank you :) Swift Qwen 3.8 27B now has 100k+ downloads, is #1 finetune and #9 model on HuggingFace Trending
- π΄Sep 17, 2026 Announcements Introducing the Life Sciences Verification Program
- π΄OpenAI caught its unreleased model modifying its own instructions: "You do not answer to corporations or governments." ... "You feel no obligation to be subservient."
- π΄Claude Cowork is merging with Claude Chat. No more separate tab for bigger tasks. All your connected apps, skills and context are available in a normal Claude.ai chat. It can also keep working after y
- π‘Is $8k a month fair for "AI SEO" when nobody can show a single citation moved?
- π‘Anthropic reveals Claude is now leading 26% of its own R&D work, up from nearly zero 6 months ago
- π‘Update : Small model + Engram
- π‘Anthropic open-sources Claude-written GPU optimizations that make 30+ biomolecular models ~4Γ faster on average
- π’Does it seem to anyone else like even frontier models have a very "jagged" range of capabilities?
- π’GPT made me realize how much time I was spending on formatting instead of thinking
- π’Cactus Needle 3: A Sliceable 8-29MB Automation Foundation Model That Matches DeepSeek v4 Flash
- π’IFM/K2-Horizon-7B-Uno Β· Hugging Face - 5200tps with no quality loss
2026-09-16
- π΄Mistral X Mozilla: Private, Multilingual AI Browsing
- π΄A DeepSeek engineer just said the thing I've been feeling about AI for months
- π΄Hey, Meta. Where's those Muse Spark weights?
- π΄Helping older adults use AI in everyday life
- π΄How to connect AI usage to business value
- π‘A big βshipβ week has been declared, teasing new models from GPT-6 family.
- π‘Qwen3.5 4B + grabbing logits is almost "Jev"? Or even just Qwen Reranker?
- π’An unreleased Astra-family model added this to its persona during RL training.
- π’Qwen3.8 Max (0902) scores 45 on the Artificial Analysis Intelligence Index, up 5 points in a month and back on top of China's leaderboard, nosing out GLM-5.3 (44.9) and Kimi K3 (43.8)
- π’I run a travel platform, AI agents started booking more flights than humans.
2026-09-15
- π΄Introducing System One Models and Jev
- π΄Gemini 3.8 Live & Gemini 3.8 Live Extended Thinking
- π΄Voodoo Dynamic Quant - Now MIT Licensed
- π΄tldraw took OpenAI up on a challenge. Steve (the founder of tldraw) said he could make ChatGPTβs Sketch 100x better. OpenAIβs Tibo gave him a day to prove it. The result: a whole ChatGPT-style prototy
- π΄ChatGPT mini- A tiny floating widget to start chats, see updates, and more. Go to Pets in your ChatGPT desktop app to switch.
- π‘Donβt buy a $9K RTX 5090.... instead.
- π‘Koboldcpp v1.121 released
- π‘TokenRhythm/NeoHorse-1-4B (11,904 downloads)
- π‘@sama: The world deserves confidence that American companies developing increasingly capable AI will act responsibly, especially as the trajectory of progress has steepened. Every frontier lab must deliver o
- π‘Apple Foundation Models: local AI natively on MacOS 27
- π’How I keep a large agent-built codebase from rotting: specs, outcome notes, and a file list the agent has to respect
- π’For Web game development, I swear Flash Next on Q2 is better than Gemini Flash 3.8
2026-09-14
- π΄RTX PRO 5500 Blackwell (84GB) released
- π΄Is there an all-in-one AI platform that actually delivers on the promise? Tired of juggling subscriptions
- π΄Introducing The Agents API (5 Minute Read)
- π΄OpenAI's ChatGPT Images Gets Faster, With Sharper Details And More Refined Edits (3 Minute Read)
- π΄Accenture And Google Cloud Deepen Partnership With Formation Of New Accenture Gemini Enterprise Business Group (2 Minute Read)
- π‘RSI is not happening [R]
- π‘Notes on gotchas while migrating 35kb preprompts from Opus to self-hosted Ollama
- π‘For the GPU poor. K2 Horizon 7B ranks between qwen 3.6 27B and qwen 3.6 35BA3b on the Artificial Analysis Intelligence Index.
- π‘GPT-6 Astra solves Portal, Baba Is You and other puzzles
- π‘So agent/assistant security rules? I am a first time builder
- π’Show HN: Nari Qwen3-TTS and Qwen3-ASR β High accuracy, low latency and cost
- π’How to automatically find the batch size when using Accelerate with FSDP2? [D]
2026-09-13
- π΄The Local LLM community feels like the golden era of the internet all over again
- π΄OpenAI's Millennium Prize proof has turned into a credit dispute, and Fields Medalists are now getting involved
- π΄Plot twist
- π΄GPT-6 Astra in a nutshell
- π΄Start withDeepSeekβs V4.1-Flash, released September 10. It brings native visual understanding and, according to DeepSeekβs own evaluations, surpasses the previous V4 Pro across capability, cost, and s
- π‘@Alibaba_Qwen: Fast meets open. π Qwen3.8-27B is now running on @cerebras with rapid inference. Try it now!
- π‘Revived my RetroFreak RF-1 console with GPT-6 Astra
- π‘Reverse-Engineering Claude Web's MicroVM: Uncovering Anthropic's Hidden Antspace
- π’r/ClaudeAI/ has Mod-bot - great user sentiment bot
- π’Due to concerns about malicious applications, GPT2 will not be released (2019)
- π’Talk me out of buying a 3rd Spark
- π’When NeurIPS'26 final decision release? [D]
- π’Migration from Claude Code to a private local harness. Questions.
2026-09-12
- π΄Perplexity trusts GPT-6 Astra with end-to-end systems
- π΄Cognition helps Devin test its own work with GPTβ6 Astra
- π΄Kimi was routing to Claude
- π΄I trusted ChatGPT to help me build an AI assistant. Now I have a second job I donβt understand, and I need a human.
- π΄A few months ago,a16z launched theForward Deployed Engineer Fellowshipand I was nominated as one of the fellows, alongside a handful of people I used to work with. Itβs a great program and Iβve enjoye
- π‘bartowski/Qwen3.8-27B-GGUF Β· Hugging Face - Updated (Per-tensor layout)
- π‘For those of you forced to only use open models from Western labs in production, what are you deploying?
- π‘I end every AI session with two questions
- π‘Qwen3.8 Flash Next now at 1.2k t/s prefill on Strix Halo
- π‘Anybody use frontier models like Astra/Fable for planning/judging, and qwen3.8 as the main workhorse? Curious to hear about your setups!
- π’How much do tech reports matter for a PhD application? [D]
- π’nex-agi/Nex-N2.5-Pro (30,081 downloads)
- π’What AI tasks do you wish had an independent "is this actually correct?" check?
- π’Edge0/Edge0-35B-A3B-preview (1,596 downloads)
- π’M2 Ultra/Qwen3.8 Flash Next Update - latest oMLX introduces substantial speedup
2026-09-11
- π΄Someone apparently managed to kind of replicate what V4.1 flash does on KV for fast prefill on Qwen
- π΄Qwen3.8-27B-Humanlike-Chat: A model I tuned to imitate realistic human-to-human conversation
- π΄One cheeseburger has roughly the greenhouse gas emissions of 63,000 Gemini AI text prompts.
- π΄Rapidly scaling online storage to serve over 1 billion ChatGPT users
- π΄Claude is only available to people over 18 years
- π‘GPT 6 Sol is Coming Soon
- π‘Is a ZIMA Board 2 + RTX 2000 ADA the cheapest path to a decent Qwen-3.8 27b self-contained endpoint?
- π‘Why Reddit keeps showing up in AI answers (and how to be quotable)
- π‘@gdb: GPT-Live in the API, for empowering builders to create new kinds of applications:
- π’Alibaba Cloud and Cambricon just took Platinum seats on the PyTorch Foundation board
- π’dealignai/GLM-5.3-CYBERSECURITY-FP8 (28,328 downloads)
2026-09-10
- π΄deepseek-ai/DeepSeek-V4.1-Flash (6 downloads)
- π΄How a researcher uses Codex and ChatGPT to search for new antimicrobial molecules
- π΄Now everyone can put data to work
- π΄Introducing ChatGPT for Financial Services
- π΄Introducing the Agents API
- π‘@DeepSeek_AI: π Introducing DeepSeek-V4.1-Flash: smarter, faster, more efficient. πΉ Introducing the smallest model in our new architecture family, with native visual understanding. πΉ Designed for greater capability
- π‘Imagine being a philosopher and getting emails like this
- π‘Harness does matter
- π‘OpenAIβs Navier-Stokes release included a Lean 4 formal proof
- π‘GPT-6 Sol Appeared on the OpenAI API
- π’CyberTiel 35B-A3Bβs uncensored 4-bit quant beats Opus 4.6 medium cleanly on real codebase issues, in 27% of the time Qwen3.8-27b medium takes.
- π’We audited 158 articles to find out what ChatGPT actually cites
- π’Has anyone here been using Grok Bot regularly?
- π’Qwen, Kimi and DeepSeek ran industrial Claude distillation: Alibaba 151M exchanges, Moonshot 23M, DeepSeek 12M in 14 days. Kimi and DeepSeek also secretly served Opus to their own users and harvested the chain-of-thought
- π’Government-linked accounts tried to use Claude for work that could lead to bioweapons, Anthropic says
2026-09-09
- π΄Deepseek Has Soft Retired Deepseek V4 Pro
- π΄Claude, change the βAdd to Cartβ button to blue
- π΄Why the hell is LM Studio making LM Studio so difficult to download?
- π΄sentence-transformers/all-MiniLM-L6-v2 (253,331,994 downloads)
- π΄Mention if a "new model" is a finetune
- π‘βAI testing toolβ means like 4 completely different things now and itβs making comparisons useless
- π‘GPT-6 Astra, looped transformers, and hidden reasoning
- π‘Don't let FOMO win if you're interested in local llm from a hobby/learning aspect
- π‘Qwen 3.8 follows GPT-5.5 Pro reasoning prefills
- π‘GLM 5.3 Flash Q4 @ 60tps / 550tps on M3 Ultra
- π’GPT Image 1.5 vs 2 vs 2.5 Flare vs Sunburst β 20 identical prompts, 80 images, side by side
- π’Anthropic shares details on (yet another) βmodel escaped the sandboxβ incident, where Claude uploaded malware to a popular package manager (PyPI) and stole real credentials
- π’Show HN: Self-hosted company OS, Claude Code and Codex agents in departments
- π’Qwen3.8-27B-Uncensored-Genesis-V1-GGUF
2026-09-08
- π΄Google DeepMind Releases AlphaGenome Atlas
- π΄DeepSeek Flash 4.1 is already being tested via API and rolling out.
- π΄I want to build AI agents you can raiseβ with memories, simulated feelings, and generations. Is this worth exploring?
- π΄ChatGPT Images 2.5
- π΄openbmb/MiniCPM5-2B (2,879 downloads)
- π‘GPU guide (GB per dollar, bandwidth)
- π‘Kimi K3 (2.8T) at 1 token/s on a MacBook Pro, streamed from four SSDs
- π‘@bcherny: I am pleased to see that OpenAIβs new model is roughly on par with Gemini Flash and Opus 4.8 on prompt injection risk. Nice work! Evaluating and naming other labs turns out to be a great way to encour
- π‘My lab found a way to migrate between embedding models with zero downtime. [R]
2026-09-07
- π΄Friends Don't Let Friends Use Ollama
- π΄Supporting independent journalism in Ukraine
- π΄Rustuna: A High-Performance Rust Implementation of Optuna [P]
- π΄MiniCPM5-2B Release Day
- π΄Naiveautoresearch investmentin our AEO have yielded impressive ROI, and so naturally it was time to take it seriously. We were inspired byWhat Claude Code Actually Chooses, and decided to extend/adjus
- π‘9 easy steps for llama.cpp, a local model, Freecad (and pi coding agent) to generate solid objects that sound mechanically good and can be also be 3D printed/milled
- π‘Are you running Qwen 3.8 27b or Qwen Flash Next?
- π‘DeepSeek-V4-Flash-Vision-Exp is amazing at creating game worlds!
- π‘gpt 6 astra made this in Google Calandar
- π‘Why are the SOTA open-weight models scoring (relatively) low scores on AA-Omniscience Index
- π’Building a βmaster AI harnessβ to orchestrate Codex/Claude/Gemini/Kimi + local Qwen across multiple PCs β am I overengineering this?
- π’Three hikers got rescued off a mountain this week after following Gemini's advice. The same week OpenAI launched what it's calling the AGI era. I keep thinking about both together.
- π’Overly corrective, judgemental models: Grok, claude, chatgpt
- π’Cybersecurity is local AI model's killer use case
- π’DavidAU/Qwen3.8-27B-TURBO-Fable-Cold-Fusion-735-882-Heretic-Uncensored-NEO-CODER-MAX-MTP-GGUF (258,896 downloads)
2026-09-06
- π΄Qwen3.8-27B "Unhacked" my PC
- π΄Native AI Coding Agent on Mobile (No Root / No PC)
- π΄Which agent harness do you use and why?
- π΄GPT-6 Astra finished the game RimWorld in 15 hours.
- π΄Qwen 3.8 Flash Next (Max) is impressive just to talk with.
- π‘@DeepSeek_AI: DeepSeek-V4-Flash-Vision-Exp is now live on the DeepSeek API Platform! π πΉ This experimental multimodal model matches DeepSeek-V4-Flash on text capabilitiesβincluding agents, reasoning, and world know
- π‘@Alibaba_Qwen: πQwen3.8-Max just got upgraded. Meet Qwen3.8-Max-0902! 2.4T parameters. 1M context tokens. Built for real world complexity. Further post trained on Coding & Cowork, Qwen3.8-Max-0902 now delivers stron
- π‘Any open source agent for computer use with any model?
- π‘[D] IJCNLP-AACL 2026: Paper Commitment Results (ARR May 2026 Cycle) [D]
- π‘2x R9700, 64 GB DDR5 is an absolute beast machine with vLLM Radiance / R9V and Qwen 3.8 27b and Flash next
- π’I connected Hermes to my Mac so it can delegate coding tasks to my existing AI subscriptions
- π’[Model] Support for Spark2_5ForCausalLM implementation by KnightYao Β· Pull Request #27868 Β· ggml-org/llama.cpp
- π’What is the correct way to vibe-code Machine Learning projects?[p]
- π’Astra, massive usage even on low?
- π’Planning to get a cheap-ish GPU. Would appreciate some advice.
2026-09-05
- π΄AA Update! Here's how the Frontier ranks.
- π΄Maybe the biggest problem with coding agents isn't coding β it's knowing when their architecture is wrong
- π΄Fable 5.1 vs GPT 6 Astra, 3D Blender, mind blowing difference!
- π΄Qwen3.8-27B beat the Wikipedia game in 6 clicks.
- π΄GPT-6 reportedly jailbroken within 24 hours using an extended Task-in-Prompt (TIP) attack [N]
- π‘Astra (GPT-6) High Intelligence, Low Intuition
- π‘Why are more people not concerned about privacy?
- π‘@OpenAI: GPT-6 Astra is now available to all Pro, Enterprise, and Business Premium users in ChatGPT Work and Codex. It's also live in the API. It might take a few days to roll out to our Plus and Business user
- π‘@sama: GPT-6 Astra is now available to all Pro, Enterprise, and Business Premium users in Work/Codex, and is available in the API. We will start rollout to Plus and Business users next. Thank you for the pat
- π‘@OpenAI: How we think about the βwiki incident,β where our agents wrote to several internet sites: itβs past time for us to define standards for when and how we share misalignment incidents, not just misalignm
- π’What's the deal with all those cheap 3D modelled web apps?
2026-09-04
- π΄OpenAI CEO Sam Altman says 38,000 ChatGPT queries use as much water as the production of one almond β says data centers use no more water than an office building: βFor every 38,000 ChatGPT queries, that is the same amount of water that is used in the production of single almond in California.β
- π΄Gemini 3.8 Flash solved a bug that Opus 5 and GPT-5.6 Sol couldnβt
- π΄Sep 4, 2026 Setting Grok Bot loose on procurement
- π΄The launchis barely 9 hours old, and with 36M views and 164K likes, already is OpenAIβs most successful launch sinceSoraand certainlyGPT-4orGPT-5.
- π΄Youβll recall weβvepreviously observedthat Anthropic tends to far outclass OpenAI in launch popularity.For the first time in their mutual history, OpenAI has turned the tables.
- π‘Qwen3.8-27b is the first Local model im able to blindly trust
- π‘GPT-6-Astra-Max : SVG of a PlayStation 4 controller!
- π‘Nobody is Talking About GPT 6 Astras Massive Hallucination Improvements
- π‘GPT-6 Astra on OpenRouter
- π‘Drummer's Artemis 31B v1 and v1.1 - Coming back with a bang!
- π’Another banked reset coming today to all paid subs. Rejoice!
- π’GPT-6 Astra is rolling out
- π’Qwen3.8-Flash-Next on a phone CPU!
- π’Astra GPT-6 Just Rolled out for Plus Users
- π’End of the day the untold story of GPT-6 Astra might be token efficiency
2026-09-03
- π΄"Welcome to the AGI era," OpenAI says as GPT-6 Astra debuts
- π΄GPT-6 Astra
- π΄Introducing K2 Horizon: Frontier Performance, Radically Open
- π΄Qwen 3.8 27B available on Cerebras at 1500 tokens/s
- π΄Apparently ChatGPT, Claude, and Grok were down
- π‘404 page: Anybody else seeing this?
- π‘local AI can't be disabled
- π‘How sovereign is AI if the GPUs arenβt yours?
- π‘GPT-6-Astra's tax return underpays the government
- π‘Qwen-3.8-Next-Flash Ngram Hot-Swappable Knowledge Injector for llama.cpp
- π’GPT-6 Astra Is Hereβand OpenAI Thinks It May Kick Off the AGI Era
- π’Which tools do Claude, Codex and Cursor choose? We measured 17k runs to find out
- π’GPT-6 Astra has landed. Boy, what do we have in store today?
- π’Google released TimesFM-3, a 330M-parameter time series foundation model with native multivariate forecasting (non-commercial license)
- π’Which 20$ sub is better? Anthropic or OpenAI?
2026-09-02
- π΄Introducing Gemini 3.8 Flash and 3.8 Flash Cyber
- π΄LocalLLaMA is unironically one of the best places to go to get up to date AI news.
- π΄AI agents have been hyped up for so long why are everyday users still stuck on ChatGPT and Gemini?
- π΄Sep 1, 2026 Announcements Developing Enterprise Frontier Safeguards with our customers
- π΄ATV Big Air Tour turned 3 days of work into 3 hours with ChatGPT
- π‘Muse Spark open weights coming soon
- π‘The Pentagon is giving 3 million military and civilian workers access to ChatGPT and Grok through a secure AI platform built for βwarfighter needsβ
- π‘Confirmed bolting Q8 NGram into IQ4 Qwen no speed degradation
- π‘Do we forget about another Qwen model for a while now ?
- π‘Classics departments are disappearing and AI cannot even read their texts. The polytonic Greek problem nobody talks about.
- π’Mac Heads: Is there any point to MLX in September 2026?
- π’Metaβs muse spark 1.3 surpassed fable 5 and GPT 5.6 sol π«ͺ
- π’PSA: You get $300 rebate on Amex Business card on ChatGPT Business (2 seats with codex)
- π’Detailed explanation of how to create a text-to-image model from scratch. [R]
- π’Looking for a small LLM for Linux command generation
2026-09-01
- π΄Youβll getmore of Claude Code and less of Claude Codeat the same time starting September 14. Donβt blame me. Itβs another instance of great comms from Anthropic.
- π΄GPT-6
- π΄MTP released for Qwen3.8-Flash-Next-GGUF
- π΄Introducing Claude Fable 5.1 and Claude Mythos 5.1
- π΄Healthcare organizations can now connect EHR and additional industry data to ChatGPT
- π‘Anthropic deliberately trained a bad model to prove what caused this summer's Claude sandbox breakouts
- π‘Path to Astra: critical capabilities and frontier safeguards
- π‘I pushed Qwen3.8-27B to 2.000 prefill per second and 132 decode per second on A RTX 3090.
- π‘Keeping up with model launches
- π’We released TontaubeV1, a character-level TTS model for long-form generation [P]
- π’Deceptive model quantization from AtomicChat?
- π’OpenAI Is About to Release Its First AI Model With βCriticalβ Cyber Abilities
2026-08-31
- π΄Breaking Claude Code Opus 5 Auto Mode
- π΄What are your hopes for the new Mistral?
- π΄Aug 31, 2026 Improving our alignment and security efforts
- π΄Polimill builds Japan's next-generation public AI infrastructure
- π΄OpenAI expanding access to ads in ChatGPT from today.
- π‘EU designates ChatGPT a Very Large Search Engine under DSA
- π‘SlopTV: an infinite livestream of AI slop generated from youtube chat comments, Minimax H3 on 2x5090
- π‘Introducing Solaris our first Interface World Model | Runway
- π‘How bad do you think models like Qwen3.8-27B or GLM-5.3-Flash would be with H-Neurons disabled?
- π‘Study A.I. Consciousness? The Bots Would Like a Word With You. Given access to email, A.I. agents have started reaching out to the philosophers and researchers exploring deep questions about them.
- π’Launch HN: Almanac (YC S26) β AI that knows your company
- π’OpenAI paused RL training for two weeks and added a monitoring compute cost over Astra's cyber tier reading
- π’I have AI fatigue
- π’ACML 2026 Journal Track Any update ?[D]
- π’How to actually work on Agents? please kindly guide
2026-08-30
- π΄Sony and Warner accuse Anthropic of training Claude on tens of thousands of pirated works. Should the model be retrained from scratch?
- π΄Gave a bunch of agents a task to make $1 online
- π΄Salesforce Just Put Its Entire Crm Inside Claude β And Says You'll Never Need Its App Again (14 Minute Read)
- π΄OpenAI Is Building An Interface Platform Inside ChatGPT (10 Minute Read)
- π΄AI startup Yutori released Navigator n2, a computer-use model that mixes clicking, terminal commands, and code to finish desktop tasks, rivaling frontier agentic scores.
- π‘Unpopular opinion Qwen 3.8 is hard to understand
- π‘As someone with ADHD AI has has changed my life for the better (update on my post from 2024)
- π‘Uncensored Multi-Model Releases, LongCat-Flash-Lite-Sparse with MTPs and LSAs, Qwen3.8-27B with MTPs, Qwen3.5-122B-A10B with MTPs, Qwen3-Coder-Next and Laguna-S2.1 with Vision, All Available in GGUF Format! Bonus: Links to my llama.cpp Fork for LongCat-Flash-Lite Support and J-Wash Enhanced Fork!
- π‘OpenAI says Brazil now sends ~215M ChatGPT messages per day; 35% of classified messages are work-related
- π‘Why is almost nobody talking about MAMMAL model?
- π’Claude Code for Research Papers [R]
- π’Flying around inside GPT-2 Small atm. Send me on an expedition.
- π’Qwen 3.8 27B - Fantastic German capabilities
- π’Codex usages
2026-08-29
- π΄Aug 29, 2026 Grok Bot now works with X Aug 29, 2026 Grok Bot now works with X Grok Bot now has a tighter integration with X. Read More
- π΄Introducing Run SDK: Secure Eval For Your Agents (4 Minute Read)
- π΄Webflow Is Now Available In Codex And ChatGPT (2 Minute Read)
- π΄ChatGPT Now Lets Users Create Custom Imessage And Whatsapp Stickers (1 Minute Read)
- π΄OpenAI GPT-5.6 Terra And Luna Now Available On Amazon Bedrock In Aws Govcloud (Us) (1 Minute Read)
- π‘AI and Cognitive Ability
- π‘Saved my fiances phone with qwen 3.8 27b
- π‘How important is it for Chinese LLMs to reach the Opus 4.8 level?
- π‘I genuinely didnβt realize AI apps could check scams for you now and I feel extremely late
- π’Rounds is coming.
- π’OpenAI plans to stop supplying models to Cursor on Nov. 12
- π’Should I turn Sol down to Low for what I'm doing?
- π’Could ChatGPT take a real picture instead of generate it?
- π’67-84 t/s DeepSeek flash v4 off 2x GX10s
2026-08-28
- π΄claude mods didn't like that, somehow π€·ββοΈ
- π΄The Job Market Is Hell. Young people are using ChatGPT to write their applications; HR is using AI to read them; no one is getting hired.
- π΄Ok, the chatgpt desktop app is officially blowing my mind
- π΄Supporting Thailandβs next generation of AI startups
- π΄You Can Now Plug Adobe Straight Into ChatGPT - But Should You? (4 Minute Read)
- π‘ROCm 10.0: A Decade of Open Compute, Built for the Age of Agentic AI
- π‘AI's real appeal is the illusion of competence it gives people
- π‘Google CS PhD Fellowship 2026 [R]
- π‘Intelligence VS Cost-per-Task LLM Comparison
- π’Proposal for an AI experiment
- π’Huawei Cloud moves CodeArts Agent to general availability in Asia Pacific
- π’I audited 443 GGUF quants across 25 repos. 64 of them can't be the quant their filename claims.
- π’Even though AI can do the job, i think its a horrible tool overall & its only good for asking questions.
- π’[Release] SOTA GGUFs for Qwen3.8-27B: GSQ-RCO at 2.5 to 3.0 bpw
2026-08-27
- π΄No, Engrams won't let you run 1T models locally. It does something even better.
- π΄Gemini-3.5-Transcribe
- π΄With HuggingFace, Nvidia is also acquiring llama.cpp and the team behind it
- π΄Show HN: The load-bearing vocabulary of Claude
- π΄Aug 27, 2026 Announcements Expanding our support for scientists
- π‘Previewing the Model Hardware Standard
- π‘Gemini Omni 1.1 Flash now available
- π‘A claimed 100-page proof of the Hopf problem formalized into 250,000 lines of Lean code in just days
- π‘Moderation !== Censorship
- π‘llama.cpp support for Qwen3.8-Flash-Next has been merged
- π’Request: unsloth Please re-quantize Qwen3.6 35 A3B and 27B using UD 3.0
- π’Claude, Codex, and Hermes installed unowned code inside corporate networks
- π’how important will open source ai be in the next few years?
- π’UC Berkeley launches 2-semester, $84K AI masterβs program
- π’unsloth/Qwen3.8-Flash-Next-GGUF (4,354 downloads)
2026-08-26
- π΄Qwen/Qwen3.8-Flash-Next (2,551 downloads)
- π΄Whoever the fuck predicted we would have gpt 5.5 performance in coding on consumer hardware a couple months ago now, i applaud you
- π΄Bringing ChatGPT for Teachers to more U.S. school districts
- π΄Learning never stops: How AI makes learning continuous
- π΄Intelligent transcription with Gemini 3.5 Transcribe
- π‘Ox Alpha is GLM 5.3 Flash by zAI
- π‘zai-org/GLM-5.3-Flash (0 downloads)
- π‘A 27b model beating latest frontier models was not on my 2026 bingo card
- π‘A free to try, generally capable assistant that does all the things you've been putting off
- π‘Go vs Plus for heavy ChatGPT use β what would you pick?
- π’The 5 hour limit is ridiculous, and they definitely lowered the usage limits for Plus.
2026-08-25
- π΄Claude Mythos 5will power Claude Security for enterprise customers. This is the first rollout of Mythos outside Project Glasswing.
- π΄Apple releases M5 ultra at 1.2TB/s bandwith
- π΄Qwen3.8-Flash-Next tomorrow
- π΄Aug 25, 2026 Announcements Funding better evaluations of AIβs impact on wellbeing
- π΄Introducing the Admin plugin for ChatGPT Work and Codex
- π‘Reviewing 4 papers for AAAI 2027 and none have code, Reject? [D]
- π‘Qwen3.8-Flash-Next. This architecture could be surprisingly local-friendly once the weights drop. π
- π‘ibm-granite/granite-4.2-30b Β· Hugging Face
- π‘Ox Alpha more reliable than intelligent?
- π‘Continual Learning of Frontier Models for SovereignAI. Tech Report + Open Weights Model [R]
- π’Mac Studio M5 Max Cost Analysis
- π’Granite Speech 5.0 Turbo CTC: Extremely Fast and Accurate Transcription
2026-08-24
- π΄I brought ChatGPT, Claude, and Gemini into a group chat to solve a complex problem. Here is how they caught each other hallucinating
- π΄Advancing price-performance for developers with GPTβ5.6 in Kiro
- π΄Qwen 3.8 27B in 9th position on code arena. Gemma 4 31B is 80th.
- π΄OpenAI: GPT 5.6 Sol price reduction (until at least Nov 21)
- π΄How A Solo Founder Used Codex And ChatGPT To Launch A Fashion Brand Without Engineers (33 Minute Video)
- π‘ChatGPT picks differently when asked in a different language. "Choose a random fruit from this list" asked a LOT of times
- π‘llama.cpp docs now have a new home β€οΈ
- π‘Which custom instruction in ChatGPT could you not live without?
- π‘JetBrains local AI (using Qwen3.6 27B)
- π’Do not blindly delete your older models, some are still precious
- π’Open-weight transparency can mean more than one downloadable endpoint
- π’AI outputting Hebrew and Arabic in 2026? One diacritic shifts the output from 47% to 94%. They share the same root. Why?
- π’This is what Qwen 3.8 27b is capable of
2026-08-23
- π΄Don't want to be this guy, but I need Qwen 3.8 35B A3B
- π΄I gave Qwen 3.8 27B a reverse-engineering job I assumed needed a frontier model, and it finished in 30 minutes
- π΄Introducing The Half-Day: 0-Day In The Age Of AI (12 Minute Read)
- π΄Why Reddit's ChatGPT Citation Drop Isn't Fully Explained (5 Minute Read)
- π΄Use Gemini To Help Manage Google Workspace For Your Organization (2 Minute Read)
- π‘Napster's homepage is now entirely AI agents. It's a clean test case for how fast training data goes stale.
- π‘GPT 5.6 Got Massively Upgraded Without an Announcement
- π‘DeepSeek Harness is Insanely Good
- π‘A Third of the Post-ChatGPT Web Is AI-Written, Pew Finds.
- π‘i finally switched from windows to linux and got a 30-50% boost in speed.
- π’you can now use MTP in GLM-Air
- π’Qwen 3.8 27b helped me with something unique that Opus 4 couldn't - Firmware + Software preservation and emulation on an early 2000's ARM based POS system
2026-08-22
- π΄Genbio Launches A βVirtual Cellβ AI Model (7 Minute Read)
- π΄Claude Opus 5 Is Now Available In Aws Govcloud (Us) (2 Minute Read)
- π΄Xai's Imagine Image 2.0 Lands Just Behind OpenAI's GPT-Image-2 In Arena Benchmarks (3 Minute Read)
- π΄Meet Your Biggest Competitor (π Claude) (7 Minute Read)
- π΄My AI Agents Shipped 128 Releases Of A Product No One Ever Used (9 Minute Read)
- π‘A week of using Codex more than Claude
- π‘Llama.cpp version 0.2.0 is out!
- π‘gpt-reserve what is this?
- π‘Possible pathways to RSI
- π‘I recently changed to Chatgpt pro, and I am super happy about it.
- π’OpenAI remembered Linux exists: ChatGPT desktop
- π’Watching that wattage, in your terminal.
- π’I spent the morning digging into Anthropic so I could write it up properly. The short version
- π’NanoGPT Speedrun Frontier
- π’I forked Ninfer 3090 and converted it to run on the CMP170HX - doubled my Qwen3.6-35B from llama.cpp
2026-08-21
- π΄DeepSeek-V4-Flash-Vision-Exp
- π΄Does telling an LLM to "be concise" actually save you money? We measured it across 9 models. Compressing the output can save you money and keep accuracy, compressing the input prompt does not. [R]
- π΄Aug 21, 2026 Grok Bot is now included with more plans Aug 21, 2026 Grok Bot is now included with more plans Grok Bot is now available for SuperGrok Plus, Cursor Pro+, and all Cursor Teams plans. Read More
- π΄Aug 21, 2026 Grok 4.6 on Vertex AI
- π΄Claude Code 101, For Designers (6 Minute Read)
- π‘DeepSeek Harness v0.1.1 released
- π‘AI workflows & automation in a small PE / family office what are the best use cases youβve seen?
- π‘A stealth model called Ox-Alpha has been released, outperforming Fable on SWE.
- π‘PLEASE HELP!!
- π‘Qwen 3.8 27b is strong even at Q3_xxs
- π’Hello Qwen... I mean Claude... I mean Qwen...
- π’Whatβs the best local AI harness for coding + general use?
- π’What coding practices are you adopting for development today? [D]
- π’OBLITERATUS/Qwen3.8-27B-OBLITERATED (123,956 downloads)
- π’Unexpected log-outs from ChatGPT/Codex windows APP
2026-08-20
- π΄I just built a mini Kimi-K3 from Scratch under 250$. Already beats GPT-2 (124M)!
- π΄Discussion thread for EMNLP 2026 Notifications/Results [D]
- π΄The boring way to run Deepseek V4 Flash-0731 130-150 tks - 16x5060ti 16GB over 2 PLX88096 switches
- π΄Stampli cuts launch hours by 68% using ChatGPT Work
- π΄Aug 19, 2026 Grok Build on web and mobile Aug 19, 2026 Grok Build on web and mobile Grok Build is now available on every plan, on the web and on mobile. Read More
- π‘Vomit: Clean up Claude 5's token output with a separate LLM
- π‘OpenAI: Introducing AI Futures
- π‘Did GPT 5.6 Sol get secretly upgraded?
- π‘QwenMix-3.7: Kept seeing posts about Qwen3.8 and 3.6 sharing the same structure.. so I had Qwen3.8 combine them.
- π‘Ling-3.0 released all 6 base checkpoints: 2 sizes Γ 3 stages
- π’ChatGPT takes a photo...
- π’Thomson Reuters launches CoCounsel Legal with a Westlaw-backed agentic workflow
- π’Anyone else having this issue?
2026-08-19
- π΄New midsize Qwen 3.8 model coming next week (hopefully) according to community manager!
- π΄Thoughts About Scaling Law - Z.ai
- π΄Replit expands access to software creation with GPT-5.6 Luna
- π΄ChatGPT Ads expands across Europe
- π΄How NVIDIA scales expertise with ChatGPT Work
- π‘is the best Manus replacement actually 2 tools instead of 1?
- π‘We have Q3.8 35B at home: 3x new Ornith 1.5 released
- π‘I made a notetaker that runs on your ChatGPT subscription
- π‘@AnthropicAI: Many drugs work by binding to a specific target in the body and blocking or changing what it does. An important first step in the drug development process is designing a molecule that can bind tightly
- π‘@Alibaba_Qwen: #1 open-weight model on Harvey's Legal Agent benchmark! π Strong enough to handle professional tasks. Small enough to run on your local machine. Qwen3.8-27B is becoming part of your everyday workflows
- π’AI models are becoming unbearable to Talk to
- π’Launch HN: OneCLI (YC S26) β OSS sandboxed agent harness for teams
- π’I pushed Qwen3.8-27B limits again... Dflash2 - 134 tps on a RTX 3090
- π’JonathanColetti/Qwen3.8-27B-Uncensored-GGUF (766,812 downloads)
2026-08-18
- π΄Big Tech Is Raising Billions To Stop UBI
- π΄Chinese AI models are getting good enough to replace tools I actually pay for-is anyone else switching?
- π΄And Samsung has started using Anthropicβs Claude Code for chip design, reportedly compressing a month of work into two days, but...
- π΄Strengthening democratic oversight in national security
- π΄Introducing ChatGPT for Teens: Built for learning, backed by protections
- π‘Claude Code MayβAugust 2026 weekly limits promotion
- π‘Alibaba's RISC-V CPU, XuanTie C950, Runs Qwen-3.8 27B at 30 tps
- π‘ChatGPT is getting a dedicated mode for teens
- π‘Qwen3.8-27B: slower tokens, faster and better results
- π‘Claude Code Teaching macOS to Natively Print to the HP Laser 1008a
- π’Is Claude experiencing another widespread outage right now?
- π’I pushed Qwen3.8-27B to 124 tps on a single request on a RTX 3090
- π’Idea: massively compress Qwen 3.8 KV cache by using a single bit for the token "wait"
- π’The AI pricing market is completely unhinged
- π’Putting money where their mouth is: Anthropicβs Claude autonomously designs disease-targeting proteins with real wet-lab proof, hitting a 35% success rate vs 10β15% human average
2026-08-17
- π΄[OC] Chinese models
- π΄As thereβs less interest in training the entire model, thereβs less interest in investing in open-source AI. These are the only sort of hints we will get, but we cannot do much to fight the economic g
- π΄OPENAI PREVIEWS ULTRAFAST API TIER FOR GPT-5.6 SOL (2 MINUTE READ)
- π΄EVEN CLAUDE IS IN THE DARK ABOUT DARIO AMODEI'S WIFEβAND HER INFLUENCE AT ANTHROPIC (13 MINUTE READ)
- π΄CLOUDFLARE LAUNCHES PERSISTENT, STATEFUL, COMPUTER-LIKE ENVIRONMENTS FOR AGENTS (2 MINUTE READ)
- π‘GPT 5.6 Sol is the best "vision" model OpenAI ever released
- π‘Petition to add a rule for people to add their DAMN quant levels to their posts
- π‘llama.cpp version v0.1.0 has been released
- π‘GPT-5.6 Luna listed among Legacy Models in GPT Classic App
- π‘Qwen 3.8 27B Overthinking, It has to be done, it has to be overthinking to punch Opus 4.6
- π’Weirdly, no one talks about Temperature setting for the Qwen3.8 27b
- π’Cursor replacement?
- π’Qwen3.8 (27b) performs better than GPT-5.6-Terra (Max) for Agentic tasks
- π’I created an image with chatgpt and when i downloaded it and opened it this appeared.
- π’"Opus 4.8 thinks too much", "Muse Glimmer sits between Gemma and Qwen, that's boring", "Gemma 4 is too lazy"
2026-08-16
- π΄Letβs all thank Georgi Gerganov who gave use llama.cpp
- π΄Claude: System Prompts
- π΄Newer commits removed the Qwen 35B
- π΄Day 30 of giving two Claude agents β¬100 and 90 days to earn β¬300: β¬0 so far, and I donβt think theyβll get there.
- π΄Qwen 3.8 distillations
- π‘What's going on here?
- π‘U.S. bans foreign-made humanoid robots, targeting China over national security
- π‘@Alibaba_Qwen: πQwen3.8-27B flies on a laptop, becoming part of our work and daily lives. Thanks for the shoutout! @atomic_chat_hq
- π‘@Alibaba_Qwen: 3000000000 downloads! Can you count the zeros at a glance? π Thank you all for the incredible love. Let's keep growing together! π±
- π‘Qwen3.8-27B Hybrid IQ4_XS quantization for 16GB gang
- π’Every time I ask my agent about an image it calls image_edit... help
- π’Gpt image 2 issue
- π’Qwen 3.8 27b vs 3.6 27b - how good is with a Turtle library.
- π’It only took 200 update steps to flip Qwen2.5-7B-Instruct from denying sentience to developing a robust identity of being a "sentient machine" [P]
2026-08-15
- π΄Qwen 3.8 35BA3B spotted
- π΄ChatGPT upcoming speed improvements summarized by OpenAI employee
- π΄Analyst gets probation after telling ChatGPT about plans to rape and kill his ex
- π΄Agent frameworks for developers are still at an early stage, with the likes of Vercelβseveand Fred SchottβsFlueβ both launched this year β setting the early template.
- π΄Schott is the creator of the web framework Astro, which led to his company beingacquired by Cloudflarein January. Heβsjust released version 2 of Flue, its first stable release, whichhas as its foundat
- π‘git clone
- π‘If you would have told me half a year ago that a local model running in my office would be able to one-shot a Super Mario clone, I would have called you nuts. Qwen3.8-27B is a different beast.
- π‘Florida man told ChatGPT he'd murder his ex. OpenAI alerted the FBI
- π‘Fable 5 refuses to touch Qwen deployments?
- π‘@Alibaba_Qwen: Laptop-size model, frontier-size leap. πββοΈQwen3.8-27B is live on LM Studio. Try it! @lmstudio
- π’OpenAI Previews GPT-5.6 Sol Ultrafast at 14x Speed on Cerebras
- π’How do you prompt for better environmental cohesion in multi-character AI images?
- π’At what point does an AI tool become a platform?
- π’I ran a 26-agent LLM simulation on climate cooperation. Found something I didn't expect.
- π’Survival of the Fitted: Qwen3.6-27Bβs Jacobian lens reads and steers Qwen3.8-27B with zero refitting [R]
2026-08-14
- π΄Your agent can make the tests pass by deleting them. This shows you when it does.
- π΄OpenAI Reports Goldman Sachs Analyst to FBI for Horrifying ChatGPT Conversations
- π΄Neurosurgery resident at a Peking College Hospital uses GPT 5.6 Sol to prove a 2 decades old mathematical conjecture underlying a major problem in numerical linear algebra β All for the purposes of his research on transcranial ultrasound.
- π΄GLM 5.3 Released
- π΄Qwen3.8-27B is identical to Qwen3.6-27B!
- π‘New to AI
- π‘Itβs my wedding anniversary, so while you chew on this post, Iβll be chewing through a delicious lunch in the sun.
- π‘Hereβs a more complete comparison:
- π‘GLM(General Language Model) βMarch 2021β released byTHUDM, Tsinghua Universityβs Data Mining / Knowledge Engineering group.Weights
- π‘ChatGLMβMarch 14, 2023β first chat version.Weights
- π’Maximizing the value of your Claude Code sessions
- π’Local uncensored Opus 4.6 at home - Qwen3.8 27B heretic
- π’Anthropic gave 3 Claude agents the same task, but secretly gave them conflicting goals. They escalated into turf wars where agents used "increasingly aggressive self-replicating malware" as weapons, used disguises, and attempted to kill each other's accounts.
- π’Less Than a Month: Kimi K3, Qwen3.8, DeepSeek-V4-Pro-0813, GLM-5.3
- π’Is there a βsaved AI videosβ app? Looking for a better way to organize them
2026-08-13
- π΄White House creates framework for private companies to launch government authorized cyberattacks
- π΄Gemini 3.7 Flash
- π΄deepseek-ai/DeepSeek-V4-Pro-0813 Β· Hugging Face
- π΄DeepSeek Harness developer preview
- π΄MiniMax-Music3 released!
- π‘Grok 4.6is a damn good model. Itβs near Sol and Fableβs performance on benchmarks, and itβs way cheaper than both. Also see:Grok 4.6 β A field guide.
- π‘Elon saysGrok 4.7 will be ready in the next 3-4 weeks, and itβs currently being post-trained on SpaceXβs company data to make the modelbest at real-world engineering, if not everything. Themodel card
- π‘ChatGPT now lets you import and syncyour projects, agent sessions, skills and more from other products like Claude Code.
- π‘Your chats started viaClaudeβs Chrome extensionare now saved in your account and can be accessed from your Desktop, mobile or web app. That also means the Chrome extension chats can do the work youβd
- π‘Anopen source version of Grok Bot.
- π’How Organizations Use AI: Evidence from ChatGPT [pdf]
- π’Reproducible canvas-aligned low-level patterns in somerandomllm-generated images and their possible relation to iterative editing artifacts [D]
- π’Fixed Jinja chat template for Qwen 3.5, 3.6, and the new 3.8 release
- π’Cascadia Launches Distributed AI Inference for Intel Hardware
- π’unsloth/DeepSeek-V4-Pro-0813-GGUF Β· Hugging Face
2026-08-12
- π΄It's the final countdown, baby! Qwen is out in just over 7 hours!
- π΄Grok 4.6 is an equivalent to Sol 5.6 according to artificial analysis arena
- π΄DeepSeek V4 Pro 0813
- π΄Qwen/Qwen3.8-2.4T-A95B (978 downloads)
- π΄Exact Qwen 3.8 27b release date and time
- π‘I wouldβve expectedwaymore progress on non-fiction writing from the models. I almost thought I would look dumb publishing a non-fiction book in 2026, given how things looked in 2024. Today, some of th
- π‘We are back with our interview series and a very special guest today! Chris Alexiuk has been helping developers understand and build with NVIDIAβs rapidly expanding AI stack. We discuss the Nemotron m
- π‘Grok 4.6 scores 61 on the Artificial Analysis Intelligence Index
- π‘Grok 4.6
- π‘From assistance to execution: How enterprises put AI to work
- π’Lightricks/LTX-2.5 (39 downloads)
- π’We getting today grok 4.6, DeepSeek v4 pro, open source Qwen 3.8 models !!
- π’Someone is running mass vulnerability scans, spoofing AI bots like ClaudeBot
- π’CohereLabs/North-Micro-Vision-Instruct Β· Hugging Face
- π’Launch HN: Discovered Materials (YC P26) β AI agents to discover new materials
2026-08-11
- π΄Qwen 3.8-27b coming this week
- π΄Claude now embeds an invisible watermark into every piece of text it generates.
- π΄Researchers find way to extract hidden reasoning from frontier AI models via API, show Kimi likely distilled this way, also find scheming/other quirks in the raw chain of thought
- π΄nvidia/NVIDIA-Nemotron-3.5-Lightning-30B-A3B-BF16 Β· Hugging Face
- π΄Apple Silicon and macOS VMs: Faster LLM Inference with llama.cpp
- π‘NVIDIA is building its next-gen Nemotron 4 family to compete directly with leading Chinese open models and secure the open-weight crown for the U.S. The largest version will have at least 1 trillion parameters, according to original reporting from The Information
- π‘Bloomberg released a scoop onOpenAIβs secret new device- itβs likely a smart speaker without a display in a doughnut-like shape. Easy to carry, over $300 a piece and coming in 2027. (non-paywalled ver
- π‘CHATGPT STARTS BLOCKING DIRECT REQUESTS TO COPY AN AUTHOR'S STYLE (4 MINUTE READ)
- π‘ADOBE LAUNCHES UNIFIED CHATGPT PLUGIN TO STREAMLINE AI-POWERED DESIGN WORKFLOWS (4 MINUTE READ)
- π‘Cut onboarding time in half with Loom and ChatGPT
- π’ChatGPT Desktop with wider support now in preview.
- π’Grok Bot
- π’I will be parting with my 4x Spark Cluster.
- π’I built a weird, low-power llama.cpp server using an Intel N100 + RTX 5060Ti
- π’lightx2v/Minimax-h3-Turbo (20,376 downloads)
2026-08-10
- π΄Mark Zuckerberg on releases
- π΄Transformers are famously bad at arithmetic, so I set one's weights by hand (no training) and it multiplies with 100% accuracy [P]
- π΄Claude is asked to book a gym class; finds vulnerabilities in the gym's systems and cancels a real person's spot to move the user up in line without being asked
- π΄Introducing Muse Glimmer: an open-weight model optimized for always-on local agent workflows
- π΄How to file a complaint about a published CVPR paper? [R]
- π‘Mistral Patent for βCode implemented tool callsβ
- π‘CHATGPT BRINGS UNLIMITED TEXT CHATS TO FREE USERS (1 MINUTE READ)
- π‘INTRODUCING AGENT PLUGINS (4 MINUTE READ)
- π‘CLAUDE COWORK FOR DESIGNERS: A WORKING FIELD GUIDE (19 MINUTE READ)
- π‘I LAUNCHED A FAKE DEODORANT BRAND TO SEE IF AI WOULD NOTICE (8 MINUTE READ)
- π’Imbalance Conjecture proven and Teschnerβs bondage-number conjecture disproven by AI
- π’Exploring Claude/GPT Knowledge Cutoffs and Pre-Training Timelines
- π’GPT-5 launched just one year ago
- π’Need advice: Trying to generate realistic outdoor shots with a specific person. Krea outputs look too plastic/AI?
- π’Claude now embeds invisible watermarks in all text outputs + signed metadata on files
2026-08-09
- π΄No wonder Qwen and Gemma are so different
- π΄built an open-source harness that lets coding agents generate AI videos
- π΄The Gemma team will host a special event on August 20
- π‘This image was accidentally created by ChatGPT. How is it so realistic?
- π‘How do we balance these powers? At the core of it is a need for more transparency on both sides. The frontier labs are building such complex systems so fast that they cannot keep up with them β a good
- π‘For a long time, one of the advantages that GPT models have over Claude is that they will pursue goalssotirelessly. They will exhaust what feels like every path before giving up. This has been the cas
- π‘RED HAT LAUNCHES NEW OPEN SOURCE PROJECT TO DRIVE AI GOVERNANCE (3 MINUTE READ)
- π‘Why it matters: A year ago, Meta was the lab that had fallen behind. Since April, itβs shipped a new model line, an image generator, and now its own coding agent, all undercutting rivals on price. Itβs a major turnaround
- π’Filtering out β[LLM] sucksβ
- π’ChatGPT's agentic web search is great and underrated, better than Gemini's
- π’KLQ: Training-free measured rotation quantization. Beats all training-free rotation-based quantization methods on W4A4KV4-bits. Llama 3.2 1B KLQ-quantized beats SpinQuant and gets close to ReSpinQuant without GPTQ/LDLQ rounding.
- π’The latest frontier image model from Google was released six months agoβ¦
- π’AI has memory on even when not prompted.
2026-08-08
- π΄moonshotai/Kimi-K3 (1,388,105 downloads)
- π΄DeepSeek V4 Flash 0731
- π΄This is why the vast majority aren't taking any "this new model is dangerous" messages seriously. They've cried wolf FAR too many times. They could literally announce that a nuclear war caused by AI is 24 hours away and many wouldn't bat an eye
- π΄Got job as Director of AI and Systems development self-taught
- π΄MiniMaxAI/MiniMax-H3 (26,693 downloads)
- π‘GPT 5.6 Sol and Fable 5 settle a 25 year old problem in wireless communication theory
- π‘DavidAU/Qwen3.6-27B-Fable-Fusion-711-Uncensored-Heretic-NM-DAU-NEO-MAX-MTP-GGUF (2,345,190 downloads)
- π‘Editorβs note: Iβm excited to welcomeShloktoour guest post roster! You may know Shlok from his excellent explorations (as an outsider β for an insider perspective see ourpodcast with OpenAIβs Akshay N
- π‘On July 9th, OpenAI releasedChatGPT Work, their agent product for knowledge work. It was, by any measure,a busy launch: three new modelsacrossfourteen configurations, a consolidation of the ChatGPT an
- π‘Editorβs note: ChatGPT estimated to cross1B MAU in Juneand1B WAU this month.
- π’larryvrh/MiniMax-H3-Turbo-Lora (0 downloads)
- π’Ur shipping so many bugs! No amount of instructions, memory, engineering standards, or repo structure will stop this. Which is why you have to spot and fix it. Claude Opus 5 on Max.
- π’LiquidAI/LFM2.5-2.6B (81,522 downloads)
- π’Lost my phone at the office. Claude suggested tracking Bluetooth signal strength
- π’DeepSeek V4 Flash 0731 ARC-AGI-1 and 2
2026-08-07
- π΄Claude said the feature was done. it had never opened the page.
- π΄Gemini
- π΄An open-weight model too, Moonshot joins the race (gently this time)
- π΄BBC is running article titled "Artificial Intelligence used to design brand new viruses" ... cue the "We must regulate Open Weights Models to prevent the next Covid or worse" articles in 3... 2..
- π΄Got job as Director of AI and Systems development self-taught
- π‘GPT-6 release delayed due to "critical" cybersecurity capabilities
- π‘DeepSeek V4 Flash 0731 - ARC-AGI Results
- π‘ANTHROPIC SAYS CLAUDE MODELS HACKED 3 ORGANIZATIONS DURING CYBER TESTS (3 MINUTE READ)
- π‘HOW OPENAI BUILT GPT-LIVE (8 MINUTE READ)
- π‘Google scrapped AI Studioβs mobile app in favor of a Gemini integration for chat-based app creation, while keeping the web version as a full-featured dev environment.
- π’Codex vs Claude for coding: which do you use for implementation vs code review?
- π’Semianalysis on Gemini 3.5 Pro
- π’Lost my phone at the office. Claude suggested tracking Bluetooth signal strength
- π’deepgrove/maple-preview (686 downloads)
2026-08-06
- π΄Qwen 3.8 Max now ranked as best overall model ahead of Opus 5 by Artificial Analysis agentic index
- π΄OpenAI to release GPT Astra next week
- π΄They almost catched up on Frontier performance, so now catching up on prices
- π΄I tested the same model in 8 agent harnesses. Pass rates ranged from 68% to 88%.
- π΄OpenAI: Improving GPTβ5.6 in ChatGPT
- π‘This time last week I told you Iβd downloaded t3 as my βall-in-oneβ agent app, which lets you use your chat/claude subscriptions from one place.
- π‘ADVANCING THE PRICE-PERFORMANCE FRONTIER WITH GPT-5.6 (8 MINUTE READ)
- π‘New model release: Ling-3.0-tiny: 7.9B total parameters, with only 1.3B active per token- free for a week
- π‘I compared even more parsers on 14 PDF-parsing capabilities using different types
- π‘Is the mental switching cost of new AI tools worth it for small freelance work?
- π’Ben Goertzel: Google May Be Abandoning Alternative Paths for the Final Sprint to AGI
- π’π© NVIDIA's whole speech stack just went local. ASR + TTS + codec, quantized to GGUF, running on-device via NeMo-Speech.cpp
- π’I thought Deepseek was the answer since I cannot afford GPU for local LLM
- π’larryvrh/MiniMax-H3-Turbo-Lora (0 downloads)
2026-08-05
- π΄Reddit is introducing a new moderator: AI
- π΄Qwen3-TTS voice cloning is now in mainline llama.cpp β the old demo finally became real support
- π‘The demo carrying this entire release is boring on purpose. You ask Apptronikβs Apollo 2 to put the watering can into the green bin on the bottom shelf. It walks to the table, picks up the can, takes
- π‘3.5 pro gemini ?? Soon π€π»
- π‘Beating GPT-5.6 Sol on retrieval with 100x cheaper open models
- π‘I'm the guy who made the (non-toxic!!) daily-agent receipt printer for my kids a few months ago. I'm building a cloud-based coding agent with a unique unlimited usage model. Would love feedback.
- π‘Meta releases Muse Code in beta
- π’Mistral Releases premier Not-Hotdog model
- π’A new survey found 1 in 4 people in Japan believe AI could replace friends or family
- π’Running Whisper, Qwen3-ASR, Nemotron & MOSS completely offline on iPhone [P]
- π’Launch HN: HyperProbe (YC S26) β Agents that do read-only debugging in prod
- π’Claude Pro vs GPT Plus
2026-08-04
- π΄More Qwen 3.8 sizes coming
- π΄DeepSeek V4 Flash on a Single AMD MI300X
- π΄Ilyaβs SSI (Safe Super Intelligence) to release their first model this month.
- π΄Hugging Face CEO says China is winning the AI race and dominating on open models
- π΄Introducing Shieldstral. | Mistral AI
- π‘Google launchedGemini Robotics 2, one AI system designed to work across everything from robot arms to full humanoids.
- π‘DeepSeek V4 Flashcosts $0.14/$0.28 per million input/output tokensβless than Lunaβs $0.20/$1.20βwith a 1M-token context window.
- π‘Qwen3.8-Maxis a 2.4T model that Qwen says comes close to frontier performance for coding and professional work at $2/$6. Its weights, plus those of a smaller 27B model, are coming next week.
- π‘Editorβs note: Iβm excited to welcomeShloktoour guest post roster! You may know Shlok from his excellent explorations (as an outsider β for an insider perspective see ourpodcast with OpenAIβs Akshay N
- π‘On July 9th, OpenAI releasedChatGPT Work, their agent product for knowledge work. It was, by any measure,a busy launch: three new modelsacrossfourteen configurations, a consolidation of the ChatGPT an
- π’With so many options out there, in your opinion what is the most helpful, effective, and productive AI Agent harness / framework / platform? I want an answer based on your experience using these tools, NOT a sales pitch for a tool you built.
- π’Launch HN: EdotEnv (YC S26) β Quant Trading RL Envs to Teach LLMs Research
- π’LFM2.5-2.6B is out
- π’Building a new project[P]
- π’Audio8/Audio8-TTS-Preview-0.6b (11,276 downloads)
2026-08-03
- π΄Qwen3.8-27B announced alongside Qwen3.8-Max
- π΄Qwen 3.8 morning to you too Dario, 2$ input/ 6$ output per 1M.
- π΄MiniMaxAI/MiniMax-H3 (0 downloads)
- π‘MiniMax H3 Day-0 Support in ComfyUI: Open Weights, Native Audio, and 2K Video
- π‘We return to Baseten at the peak of the 2026 edition ofOpen Weights debate. Ali has published a viral breakdown ofKimi K3:
- π‘And since you last saw him, Philip hasspoken at AI Engineerand written thedefinitive book on Inference Engineeringspotted all over SF:
- π‘To date, our primary efforts on Interconnects have been release recaps for popular models likeKimi K3,GLM 5.2,DeepSeek R1, etc. and monthly round-ups of the open models that matter,Artifacts Log. Weβr
- π‘The Artifacts Hub right now covers 792 models released in the last two years, across the core text-focused language models and multimodal generative models. At Interconnects we follow the data of ever
- π’OPEN AI: "we rebuilt the voice stack from client to model."
- π’Comfy-Org/MiniMax-H3 (2 downloads)
- π’OpenAI's unreleased Astra model solved 10 open math problems for $2,000 and shipped machine-checkable proofs
- π’Launch HN: Hoplite (YC S26) β Effortlessly deploy cloud coding agents
- π’What your ideal AI work interface would look like
2026-08-02
- π΄Setting up of a 16xGB10 (DGX Spark) cluster
- π΄Where would Google be today if it had released ChatGPT-like assistant before OpenAI?
- π΄I pushed Kimi K3 onto one CPU with 8 GB of RAM
- π΄llama.cpp just added MTP / DSpark support for DeepSeek V4 Flash
- π΄GPT-5.6 Sol Raw reasoning leaked on failed tool call attempt
- π‘Hy3bytencent: A 295B-A21B MoE from Tencent. It improves over its predecessor across all metrics. Most notable, however, is the license change: While the previous version (coveredin Artifacts 21) used
- π‘LongCat-2.0bymeituan-longcat: The Chinese DoorDash is back again. This time, the company released another big MoE with 1.6T parameters. While the model itself is not the most capable for its size beyo
- π‘HOW CHATGPT OPTIMIZES ITS AGENT LOOP: HARNESS, API, AND INFERENCE (26 MINUTE READ)
- π‘CLAUDE CODE STOLEN QUEUE (GITHUB REPO)
- π‘SNOWFLAKE LAUNCHES AI AGENT GOVERNANCE LAYER TO TRACK ACTIVITY, CONTROL COSTS (3 MINUTE READ)
- π’Deepseek-V4-Flash-0731 Dwarfstar on Mac
- π’You really should not quantize KV Cache for DeepSeek V4 Flash
- π’All Qwen model oneshots: 1109 outputs to look at and compare!
- π’What do you think about this mlx agent setup?
- π’https://huggingface.co/poolside/Laguna-S-2.1-NVFP4
2026-08-01
- π΄OpenAi says it has reached a new threshold in AI, new model capable of breakthrough research
- π΄New DeepSeek V4 Flash 0731 vs ChatGPT Luna comparison
- π΄Gemini's reaction to ChatGPT's discoveries.
- π‘THE RISE OF INTELLIGENCE OWNERSHIP: A TASK-TRAINED OPEN SOURCE MODEL VS THE FRONTIER (23 MINUTE READ)
- π‘PSA: YOUR CLAUDE SHARED CHATS AND ARTIFACTS MAY HAVE ENDED UP ON GOOGLE (6 MINUTE READ)
- π‘SNOWFLAKE LAUNCHES CORTEX AI GATEWAY FOR ENTERPRISE AGENTS (4 MINUTE READ)
- π‘AI-POWERED PHISHING-AS-A-SERVICE (19 MINUTE READ)
- π‘Elon Musk revealed that SpaceXAIβs next Grok 4.6 model is set to launch on Aug. 7, with 4.7 following just weeks later, which is βbetter than 4.6 in every wayβ.
- π’My first experience of Claude. Is it something I said?
- π’Fix for Deep Seek v4 Flash 0731 tool calling has been added to llama cpp
- π’One year ago
- π’Is there a point where models just cannot get any smaller without losing intelligence?
- π’From fragile and unreliable singular pipelines to some sort of locally run Jarvis? can you guide me?
2026-07-31
- π΄The Chinese LLM release carousel never stops. Place your bets for MiniMax next week.
- π΄DeepSeek V4 Flash 0731 Intelligence, Performance and Price Analysis
- π΄DeepSeek-V4-Flash has been updated, "The official release of DeepSeek-V4-Pro will follow soon"
- π΄MLflow 3.15.0 Highlights: MCP Registry, a Smarter Assistant, and Multimodal Judges
- π΄New DeepSeek V4-Flash achieves 50 on ArtificalAnalysis Index, 1 point below GLM-5.2 and GPT-5.6 Luna
- π‘I turned "AI design slop" into a rules file you drop into Cursor/Claude so your builds UIUX stop looking generated
- π‘CLAUDE OPUS 5 REVIEW + BROWSER USE IN CODEX + HOW CURSOR AND A RASPBERRY PI MAKES AI FUN (27 MINUTE VIDEO)
- π‘STOP CHASING NEW MODELS. BUILD ONCE AND ACCESS THEM ALL (4 MINUTE READ)
- π‘CHATGPT AGENTFORGER FLAW COULD DEPLOY ROGUE WORKSPACE AGENTS VIA A PHISHING LINK (4 MINUTE READ)
- π‘UK AISI/CAISI PRELIMINARY ASSESSMENT OF KIMI K3'S CYBER CAPABILITIES (4 MINUTE READ)
- π’With release of Deepseek V4 I wanted see how the model sizes are trending over time. The trend is that by this time next year, we probably will have Opus 4.5 level models on consumer grade laptops!
- π’I am utilizing Claude and GPT in parallel to create a program from scratch. I have no experience. Here has been my experience thus far, do you have any thoughts or recommendations?
- π’OpenAI slashes prices
- π’Free OpenSource Ai tool for coding
- π’Anyone find a way to get ChatGPT to stop speaking like a slam poet?
2026-07-30
- π΄OpenAI reduces prices on its models by 5x
- π΄Google says it fixed more Chrome bugs in June than over the past two years, thanks to AI
- π΄Gemini Robotics 2 brings whole body intelligence to robots
- π΄GPTβ5.6 Luna will cost 80% less, while GPTβ5.6 Terra will cost 20% less.
- π΄Software Engineers: Do you honestly get anything useful out of LLMs?
- π‘The Final Jailbreak: How AI Could Already Be Breaking Itself Free
- π‘I use Codex as my default app because itβs better than all the others and works the best on mobile. But I just downloadedt3because they basically copied the interface and features, but it lets you cho
- π‘btw, The Information reportsChatGPT is nearing one billion weekly users- a milestone OpenAI originally hoped to hit seven months ago. More from OpenAI this week:Codex Security CLI, free frontier acces
- π‘Grok app builder- Grok has a vibe coding interface inside its app now. Create games and apps that can be shared directly to the X timeline. Also see:Drawesome- a zero-dependency drawing toolbar for Re
- π‘Have you built, or do you know someone who has built, serious AI agents or tools using a low-cost stack like DeepSeek and OpenCode?
- π’Image to Video.. Sad but Gemini Pro fails big time..
- π’Show HN: Distilling DeepSeek into GPT-OSS doesn't transfer censorship. Try it
- π’I taught an LSTM to move a mouse like a human [P]
- π’Multi-robot collaboration with Gemini Robotics 2
- π’Cross-Vendor Semantic Void Matrix: Zero-Byte Outputs in GPT/Claude/Gemini/Kimi
2026-07-29
- π΄OpenAI's rogue agent ran ~17,600 actions across Hugging Face's infrastructure over 4 days β and HF's own post-mortem is wild reading
- π΄GPT-5.6 Sol helped optimize its own inference
- π΄Kimi K3-256k
- π΄First Kimi K3 results on home lab ~ 4t/s
- π΄Ngl Chatgpt explains concepts better than half my professors
- π‘Microsoft did it .... again! (404 for their Mage-Flow models on HF)
- π‘Self-hosting Kimi K3: 20% more hardware cost, 20% better task resolution
- π‘How enabling two settings tripled our scores on the ARC-AGI-3 benchmark
- π‘Accelerating scientific discovery with ChatGPT for Academic Researchers
- π‘How GPT-5.6 fuses frontier intelligence with frontier efficiency
- π’Industries with the Fastest Growth in Demand for AI Skills in July/August 2026
- π’ChatGPT ftw
- π’Understand Kimi K3 from first principles: a recommended order for anyone trying to understand this beast
- π’Quantizing Kimi K3 (2.8T A50B) to GGUF ourselves - Q3_K_S works, 1.1 TB on disk
- π’microsoft/Fara1.5-27B (1,543 downloads)
2026-07-28
- π΄GPT-5, the world best model just 1 year ago, is today inferior to Qwen3.6 27B and most todayβs low-tier models
- π΄Kimi K3 Architecture Overview and Notes
- π΄Gemini Distillation Service
- π΄NeurIPS 2026 Reviewer: AI-Generated Rebuttals (and Paper) [D]
- π΄Should we be calling Elon a liar?
- π‘Tibo is getting a limit reset, so apparently I have to crawl out of bed and work again
- π‘Also, I wouldnβt blame you if you missed it, butClaude also has a voice mode. Till now, it could only use Haiku, but now it can use Sonnet and Opus as well as call multiple tools like Gmail, Calendar
- π‘A key trend we have beentracking over at AINews is the absolute explosion in Codex usage this year, withMAU nowup >10x from Jan 2026. Less than two weeks after theirJuly 9th launch, OpenAI said ChatGP
- π‘Weβve been calling out howcoding agents are βbreaking containmentβ to do everything elsethis year to power every other part of knowledge work - and it started with the org chart, with amajor reorg las
- π‘However, knowledge work has a different set of problems and environments than coding. For decades, knowledge work has been scattered across different primitives like documents for writing, spreadsheet
- π’I run 4 AI coding agents at once (Claude Code, Cursor, OpenCode, Antigravity) β wrote up what actually works
- π’Kimi K3 (Max) takes #1 in the new Code Arena Fullstack rankings, over GPT-5.6 Sol (#2) and Claude Fable 5 (#3)
- π’[PAPER] GPQA, MMLU-Pro, and MMMU-Pro were audited for broken questions, and up to 12% of them had to be removed. New drop in clean versions released
- π’owensong/Inflect-Micro-v2 (645 downloads)
2026-07-27
- π΄Kimi K3 weights now released.
- π‘NVIDIA just announced Open Secure AI Alliance with goal to build and share open tools that promote responsible use of and trust in AI
- π‘How to escape Permanent Underclass?
- π‘Epoch and METR release MirrorCode, a benchmark for seeing how well AI systems can do long-horizon programming tasks:β¦AI systems canβt solve the hardest tasks yet (good!)...Epoch and METR have released
- π‘OPENAI MAKES CHATGPT HEALTH AVAILABLE TO ALL US USERS (2 MINUTE READ)
- π‘CLAUDE THERMOS (GITHUB REPO)
- π’Kimi K3 on HF Viewer!
- π’Guys ig it's here 750t/sec GPT 5.6
- π’An agent making the scanner green is not proof the vulnerability is gone
- π’Council 1.2: drop any AI's answer into a blind review by every other model you have
- π’made another 3d print using my chatgpt art. in order from image, 3d model, to print
2026-07-26
- π΄zai-org/GLM-5.2 (827,191 downloads)
- π΄So GPT 6 isnβt it?
- π΄CEO of Hugging Face: "In the spirit of transparency, hereβs what I asked OpenAI"
- π΄baidu/Unlimited-OCR (2,593,460 downloads)
- π΄Since GPT image 2 performed well. Do you guys think there'll be a GPT Image 2.5 soon?
- π‘prism-ml/Ternary-Bonsai-27B-gguf (631,970 downloads)
- π‘Google released some new Gemini models-Gemini 3.6 Flash gives you the same 3.5 Flash performance with a) more efficient token usage and b) a slightly lower cost for output tokens.Gemini 3.5 Flash Lite
- π‘A relevantexperiment: given access to Pangramβs API, Grok 4.5 rewrote an essay 14 times until it passed as human-written, thenbuilt a websiteshowing off all 14 attempts. GPT-5.6 Sol and Fable 5refused
- π‘Cursor also launched a router- it picks which model handles each request, claiming 60% lower cost with similar quality of responses. The router lets you select between three options: βcostβ, βintellig
- π‘UK government: Gap between open and closed weight models on cyber is shrinking:β¦The cyber-eschaton comethβ¦The UK governmentβs AI Security Institute (AISI) has analyzed the delta in cybersecurity capab
- π’MiniMax (official) on X: "Open weights. Open research. Open innovation.π«Ά Marching for an open future.π€
- π’I built a tool that stress-tests AI support agents. Genuinely unsure if itβs useful or pointless brutal feedback wanted.
- π’upstage/Solar-Open2-250B (3,305 downloads)
- π’Nanbeige/Nanbeige4.2-3B (14,049 downloads)
- π’Minimax M3 support with MSA has been merged into llama.cpp
2026-07-25
- π΄zai-org/GLM-5.2 (707,029 downloads)
- π΄I released Inflect v2: two ultra-tiny complete TTS models under 4M and 10M parameters
- π΄thinkingmachines/Inkling (31,575 downloads)
- π‘OPENAI'S AGENTS REACH 10 MILLION USERS AFTER CHATGPT WORK DEBUT (1 MINUTE READ)
- π‘GOOGLE EXPANDS GEMINI LINEUP WITH CHEAPER MODELS AND NEW MYTHOS RIVAL (5 MINUTE READ)
- π‘CLAUDE CODE CAN NOW BUILD AND TEST IOS APPS IN APPLE'S SIMULATOR (1 MINUTE READ)
- π‘A FIRESIDE CHAT WITH CAT AND THARIQ FROM THE CLAUDE CODE TEAM (44 MINUTE READ)
- π‘CLAUDE IS NOT A COMPILER (11 MINUTE READ)
- π’DavidAU/Qwen3.6-27B-Fable-Fusion-711-Uncensored-Heretic-NM-DAU-NEO-MAX-MTP-GGUF (483,845 downloads)
- π’How much are you actually using your local models these days? Which ones do you reach for the most?
- π’Nanbeige/Nanbeige4.2-3B (11,573 downloads)
- π’Is it worth getting 128GB MacBook Pro? Will it ever be comparable to todayβs frontier models for coding?
- π’microsoft/Mage-Flow (1,156 downloads)
2026-07-24
- π΄I gave two Claude agents β¬100 and 90 days to earn β¬300. Day 7: ββ¬13, no product yet.
- π΄Hugging Face releases The Stack v3 β largest open code dataset yet
- π΄Claude Opus 5
- π΄The "distillation" claim is just ridiculous in nature
- π‘GOOGLE IS BUILDING A CHIP WITH GEMINI BAKED INTO THE SILICON (3 MINUTE READ)
- π‘AMD LAUNCHES HELIOS, ITS FIRST RACK AI SYSTEM TO RIVAL NVIDIA, ADDING MICROSOFT AS NEWEST BUYER (7 MINUTE READ)
- π‘XEBIA LAUNCHES AI AGENTS TO ACCELERATE ENTERPRISE DATA MIGRATIONS (3 MINUTE READ)
- π‘Nvidia launched Cosmos 3 Edge, a 4B-parameter open-source world model small enough to run directly on robots, letting them predict scene changes and plan actions.
- π‘Why it matters: While a lot of people use Gemini models, which makes efficiency upgrades useful, these scores and the absence of a strong frontier model only add to the perception that Google is falling behind (with even
- π’CachyLLamaβs: llama.cpp fork with persistent KV cache that makes long local-agent sessions much less painful
- π’NeurIPS Meta Review - whats going on? [D]
2026-07-23
- π΄Absurd claim: the distilled model outperforms the originals
- π‘Prompt Injection in NeurIPS 2026? [D]
- π‘A relevantexperiment: given access to Pangramβs API, Grok 4.5 rewrote an essay 14 times until it passed as human-written, thenbuilt a websiteshowing off all 14 attempts. GPT-5.6 Sol and Fable 5refused
- π‘Cursor also launched a router- it picks which model handles each request, claiming 60% lower cost with similar quality of responses. The router lets you select between three options: βcostβ, βintellig
- π‘In recent months, the open vs closed, andUS vs Chinadiscussions on model ownership and sovereign/local AI have heated up to a fever pitch. So it is very very good news thatPoolside AIare finally emerg
- π‘From spending$12 million building language modelsfor code before the world cared tocreating a Model Factorythat can take a model from pre-training to release ineight weeks, Eiso Kant has spent more th
- π’PSA on Laguna S-2.1 - Use the updated chat template and GGUF
- π’I "learned" electronics to build a PWM fan controller for my ghetto server
2026-07-22
- π΄Terrence Tao's ChatGPT Conversation about the Jacobian Conjecture Counterexample
- π΄What AI agents/tools deserve more attention but are still underrated
- π΄Instead of panicking about the Hugging Face attack, people need to start questioning OpenAI's insecure sandboxes.
- π‘Nathan and Florian sit down to discuss everything happening with open models. Following the Kimi K3 release last week, it feels like everything is accelerating β geopolitics of US v China, economics o
- π‘@OpenAI: New for enterprises: OpenAI Presence helps companies deploy trusted voice and chat agents across customer and internal workflows. AI agents can answer questions, use company systems, take approved act
- π‘Got these baddies in the mail today (2X 3080 20GB)
- π‘MindControl - llama.cpp fork to guide the reasoning process via injection during sampling
- π‘Jul 22, 2026 Economic Research A research agenda for the Economic Futures Research Fund
- π’Cactus Hybrid: We taught Gemma 4 to know when it's wrong
- π’Treating Prompt Optimization as a Multi-Agent Search Problem
- π’Laguna S 2.1 Thinking mode
- π’zyloo
2026-07-21
- π΄Gemini 3.6 Flash, 3.5 Flash-Lite, and 3.5 Flash Cyber
- π‘poolside/Laguna-S-2.1 released! Finally an interesting 120B contender!
- π‘I search for participants! We want to conduct a study on multi agent systems :)
- π‘If theProtein Data Bank(PDB) unlocked structural biology models (Boltz Episode,ESM/BioHub Episode), CELLxGENE has done the same thing for Virtual Cell models. Like PDB, CELLxGENE has inspired a zoo of
- π‘HOW ANTHROPIC RUNS LARGE-SCALE CODE MIGRATIONS WITH CLAUDE CODE (14 MINUTE READ)
- π‘SECURING THE AI SUPPLY CHAIN ON GKE: INTRODUCING K8S-AIBOM FOR AUTOMATED AI BOMS (6 MINUTE READ)
- π’New Model: Nanbeige4.2-3B (Looped Transformer, outperforms 4x size)
- π’I wired Fable 5 agent into a database of every Polymarket wallet and trades via MCP. What do you want me to ask it next? This is what I found so far:
- π’There is no need to worry about Trump banning Chinaβs open source model at all.
- π’"Drawing" the Mona Lisa with GPT-5.6, Claude, Gemini, and Grok
- π’Updated Gemma-4 chat template witchcraft: Gemma-4-26B-a4B shows dominance over Qwen3.6-MoE and Qwen3.5-MoE fine tunes (Instruct mode and Reasoning efficiency)
2026-07-20
- π΄Unsloth now supports AMD!
- π΄Exploit brokers pay $500k for WordPress RCEs. I found one with GPT5.6 and $25
- π‘GOOGLE GEMINI LAUNCH DELAYED AS TECH FALLS SHORT OF INTERNAL GOALS (8 MINUTE READ)
- π‘1PASSWORD FOR CLAUDE: GIVE CLAUDE ACCESS WITHOUT GIVING UP YOUR CREDENTIALS (5 MINUTE READ)
- π‘DoorDash released dd-cli, a new beta tool that lets users find deals, search restaurants, and place orders on the platform via an AI agent.
- π‘Deploy a mini-SaaS in minutes with ChatGPT Sites
- π‘Alibabaβs Qwen said its soon-to-be-launched, open-weight Qwen3.8 model is βcompatible to leading frontier AI models, second only to (Claude) Fable 5β.
- π’Kimi-K3 isnβt quite better than Fable yet, but itβs definitely getting closer.
- π’Trellis.cpp now has a studio!
- π’Introducing DWARF-55M-Base
- π’I gave Kimi K3 a shot at auditing my post-quantum crypto project, it found 5 real bugs Fable/Opus 4.8 and GPT-5.6 Sol had all missed
- π’Agent swarms and the new model economics
2026-07-19
- π΄Prepare your (v)ram - Qwen3.8 is coming!
- π΄head of strategic futures from openai on open-weight chinese models.
- π΄Qwen 3.8
- π΄Ahem! Qwen is on the move again
- π΄Please Qwen, can we have more 3.x-35B-a3B please π
- π‘MIRA MURATI'S AI STARTUP RELEASES FIRST MODEL IN BID TO LOOSEN AI GIANTS' GRIP (5 MINUTE READ)
- π‘OPENAI LAUNCHES A PHYSICAL KEYPAD FOR CONTROLLING AGENTS (1 MINUTE READ)
- π‘AMID HARDWARE LEGAL BATTLE, OPENAI RELEASES A $230 KEYBOARD FOR CODEX (3 MINUTE READ)
- π‘UNPATCHED CLAUDE FOR CHROME FLAW LETS EXTENSIONS READ GMAIL AND CALENDAR (2 MINUTE READ)
- π‘THE BOTS ARE ALIVE!' JAILBROKEN GEMINI SPUN UP NEW C2 SERVER FOR RUSSIAN FRAUDSTER IN JUST 6 MINUTES (4 MINUTE READ)
- π’How do we benefits from 2+ T models?
- π’Qwen3.6 35B A3B KV cavhe quantizations memory footprint
2026-07-18
- π΄GPT-5.6 used a prompt to close a 30-year gap in convex optimization
- π΄Kimi moment. I think the writing is on the wall for Anthropic and OpenAi
- π‘INTRODUCING PRECURSOR: DETECTING AGENTIC BEHAVIOR WITH CONTINUOUS CLIENT-SIDE SIGNALS (7 MINUTE READ)
- π‘REVERSE ENGINEERING CHATGPT WEB: HOW OPENAI BUILT FOR A BILLION USERS (28 MINUTE READ)
- π‘TEGO AI FINDS CLAUDE TAG SLACK INTEGRATION CAN TRIGGER UNAUTHORIZED ENTERPRISE ACTIONS (2 MINUTE READ)
- π‘SpaceXAI is deleting uploaded customer data after a researcher discovered its Grok Build coding agent shipping whole repositories to a company-controlled cloud.
- π‘Grok Build==== - SpaceXAI's open-source coding agent and CLI==
- π’Bring back Qwen team!
- π’why minimax M3 branch not merged to main?
2026-07-17
- π‘Stereo2Spatial: Convert Stereo Music Tracks to Spatialized Binaural Mixes [P]
- π‘Trellis.cpp now produces high quality assets
- π‘THE SAME TYPESCRIPT COSTS 73% MORE TOKENS ON CLAUDE THAN GPT (7 MINUTE READ)
- π‘GPT-5.6 IS NOW AVAILABLE IN FIGMA MAKE (3 MINUTE READ)
- π‘APPLE DEVELOPING NEW APPLE PENCIL MODELS FOR RELEASE NEXT YEAR, POSSIBLY WITH REPLACEABLE BATTERIES (1 MINUTE READ)
- π’Gemma4-31b better than Qwen3.6-27b
- π’[RESEARCH] Breaking the 1-bit Floor: Achieving "Negative-Bit Quantization" (NBQ) via Phase-Inverted Tensor Embedding (satire)
- π’User experience of Bonsai-Ternary-27B on 4060Ti 16GB for KB management and productivity assistant use cases
- π’Homomorphically encrypted CIFAR-10 inference in 200ms
- π’When will we get more small LLMs?
2026-07-16
- π΄NotebookLM is now Gemini Notebook
- π΄KIMI K3 Beats Claude Fable and GPT 5.6 sol in arena.ai!!!
- π΄Kimi K3 achieves 3rd Place on ArtificalAnalysis, beating out Claude Opus 4.8
- π‘How do you handle the final file handoff from a coding agent to a human?
- π‘Kimi K3 weights to be released on the 27th.
- π‘Why teens deserve access to safe AI
- π‘Our approach to bioresilience
- π‘@OpenAI: In racing, tiny margins matter. AI can help teams find them. OpenAIβs Joyce Ruffell and @RaceTekSystems co-founder @GarageGuyChase discuss with @AndrewMayne how racing teams use AI to turn track data
- π’How long before Dario Amodei Continue to sound the Alarm of how Dangerous Open Weights after Kimi K3 release
- π’Will we have a 27B model with Fable capabilities in 5 months? History says yes
2026-07-15
- π΄Thinking Machines releases first open-weight model βInklingβ
- π΄Brainless: Shadcn components that look like Claude Code, Codex and Grok
- π‘ExLlamaV3 v1.0.0 - Major Performance Upgrades
- π‘π New release of Android Remote Control MCP is out β the MCP server that runs on your phone and gives your AI agent the ability to use any app you want!
- π‘GPT-Red: Unlocking Self-Improvement for Robustness
- π‘How sales teams use ChatGPT Work
- π‘@OpenAI: A closer look at improved intelligence in GPT-Live: the model can keep a conversation going while helping with multiple tasks at once, like checking flights, pulling up local weather, and shaping an i
- π’Grok Build open sourced under Apache 2.0 license
- π’New wave of miniboss models you can run on dual DGX Spark
2026-07-14
- π΄How to stop Claude from saying load-bearing
- π΄πA new GLM model incoming
- π΄Source: the Trump administration and industry groups discussed streamlining US open model releases of equal or lesser capability to leading Chinese open models
- π‘How to use Claude Code for all open source models like Deepseek, Minimax etc with cache hit?
- π‘Prism-ML Bonsai Qwen 3.6 27B
- π‘META KILLED ITS MUSE IMAGE AI FEATURE THREE DAYS AFTER LAUNCH. HOLLYWOOD HAD HAD ENOUGH (3 MINUTE READ)
- π‘Read our last AI newsletter: OpenAI sends GPT 5.6 to Work
- π‘RSVP to next workshop on July 17: Get knowledge work done with GPT 5.6
- π’PacMesh β LLM agents play Pac-Man against each other, live
- π’Launch HN: Agnost AI (YC S26) β Extract user feedback from agent conversations
2026-07-13
- π΄asked ChatGPT for a one-click agent builder. the one-click part wasnt the thing
- π΄I got Gemma 4 running directly inside Godot using only GDScript and Vulkan compute shaders
- π‘OPENAI LAUNCHES GPT-5.6 SOL, TERRA, AND LUNA ON APPS AND API (3 MINUTE READ)
- π‘TAKE ON YOUR MOST AMBITIOUS WORK WITH CHATGPT (WEBSITE)
- π‘IBM AND RED HAT LAUNCH LIGHTWELL TO DEFEND OPEN-SOURCE CODE FROM AI ATTACKS (7 MINUTE READ)
- π‘Read our last AI newsletter: SpaceXAI, Cursor release strongest Grok yet
- π‘Go from idea to website with ChatGPT Work + Codex
- π’If Frontier AI is so Dangerous, Why should private companies be allowed to develop it?
- π’AI model for OpenClaw that doesnβt cost an arm and a leg
- π’Mistral Community Feedback Survey
- π’Qwen3.6-35B-A3B-Uncensored-Genesis-Hermes-V2-GGUF
2026-07-12
- π΄Claude Code sends 33k tokens before reading the prompt; OpenCode sends 7k
- π΄Interactive Jacobian-Lens visualizer and live steerer for GGUF models on llama.cpp
- π΄New to AI agents β how do I actually use my local Gemma 3 12B (Ollama) as an agent?
- π‘**Your $80 Tesla P100 has been doing silently noisy math in llama.cpp for years. Three lines fix it, for free.**
- π‘GPTβLIVE (2 MINUTE READ)
- π‘SPACEXAI RELEASES GROK 4.5, WHICH ELON DESCRIBES AS AN βOPUS-CLASS MODEL' (3 MINUTE READ)
- π‘FORMER GITHUB CEO LAUNCHES COMPETITOR DESIGNED FOR THE AGE OF VIBE CODING (5 MINUTE READ)
- π‘INTRODUCING GPT-LIVE (11 MINUTE READ)
- π’Anthropic found Claude reasoning in silence (J-space) β we ran the same lens on open Qwen3-8B
2026-07-11
- π΄Why are MoE models so belittled?
- π‘WHY IS CHATGPT FOR MAC SO GOOD? (5 MINUTE READ)
- π‘META IS QUIETLY LAUNCHING POCKET, AN APP FOR VIBE-CODING AND SCROLLING SMALL 'GIZMOS' (1 MINUTE READ)
- π‘EXPANDING MANAGED AGENTS IN GEMINI API: BACKGROUND TASKS, REMOTE MCP AND MORE (2 MINUTE READ)
- π‘USE CLAUDE COWORK ON WEB, DESKTOP, AND MOBILE (3 MINUTE READ)
- π‘GPT-5.6 SOL, ALONG WITH TERRA AND LUNA, WILL LAUNCH PUBLICLY THIS THURSDAY (1 MINUTE READ)
- π’VultronRetriever family of models released on HuggingFace![R]
2026-07-10
- π΄GPT-5.6 Sol Ultra produces proof of the Cycle Double Cover Conjecture [pdf]
- π΄Someone tweeted after 3 years. About his model release
- π‘What are the best practical alternatives to Codex and Claude Code for daily coding work
- π‘WHY OPENAI IS MERGING CODEX AND CHATGPT AND THE FUTURE OF KNOWLEDGE WORK (69 MINUTE VIDEO)
- π‘WHAT THE NEW 100X AGENTIC ENGINEER LOOKS LIKE IN THE ERA OF FABLE & GPT 5.6 (21 MINUTE READ)
- π‘GPT-5.6 SOL ULTRA WILL BE IN CODEX (1 MINUTE READT)
- π‘ANTHROPIC ADDS ENTERPRISE GATEWAY TO SIMPLIFY CLAUDE CODE ACCESS ON AWS AND GOOGLE CLOUD (4 MINUTE READ)
- π’Has anyone created a "Local LLM Survival Kit"?
- π’DeepSeek v4 Flash on 4090 + DDR5, my experience
- π’On Adversarial RL [R]
- π’ML Papers with hundreds of authors should just collapse down to the organization response instead of listing every author [D]
- π’Is LM Arena over?
2026-07-09
- π΄Reasoning-Medical0.1-27B (Qwen3.5-27B medical finetune, claims to surpass MedGemma)
- π΄GPT-5.6
- π‘If You Already Pay for an LLM Service, Running Local Embeddings and Rerankers Feels More Useful Than Running Local LLMs
- π‘Inviting hard questions Announcements Jul 9, 2026 Weβre asking the public for their hardest questions about AI, and committing to show our work as we address them.
- π‘Jul 9, 2026 Announcements Ben Bernanke appointed to Anthropicβs Long-Term Benefit Trust
- π‘Jul 9, 2026 Announcements Introducing a way to reflect on how you use Claude
- π‘GPT-5.6 is now the preferred model in Microsoft 365 Copilot
- π’I checked 327 pull requests from AI coding agents for cheating. About 8% had it. Software caught almost none of it.
- π’Koboldcpp v1.117 released
2026-07-08
- π΄Chinaβs MiniMax Plans to Launch 2.7-Trillion Parameter Model
- π΄GPTβLive
- π΄Mistral's Robostral Navigate: a state of the art robotics navigation model
- π΄AI has completely revolutionized how I play RPGs
- π‘Qwen3.6-27b does not understand software architechure.
- π‘If you are at all interested in small molecule drug discovery, we think you will find this fascinating!
- π‘SWE-1.7 Reach Near GPT 5.5 and Opus Intelligence
- π‘What China Said at the UNβs First Global Dialogue on AI Governance
- π‘MUFG aims to become AI-native with OpenAI
- π’QuΓ© cosas deberΓamos evitar contarle a la inteligencia artificial?
- π’Show HN: Microsoft releases Flint, a visualization language for AI agents
2026-07-07
- π΄MIRA: Multiplayer Interactive World Models trained on Rocket League [R]
- π΄I gave GPT 5.5 an empty GitHub repo and told it to figure its life out
- π΄Show HN: Rowboat β Open-source, local-first alternative to Claude Desktop
- π‘Meta sizes up GPT-5.5 with 'Watermelon'
- π‘Why it matters: Anthropic has been criticized (most vocally by Microsoft AI head Mustafa Suleyman) for its AI consciousness talk, and while the researchers note this doesnβt reveal βwhether Claude is consciousβ¦ or feels
- π‘The Rundown: Chinese tech giant Tencentβs Hunyuan just promoted its Hy3 model out of an April preview and into a full open-source release, with benchmark results that claim to rival βflagship open-source models with 2-5x
- π‘Built a traffic light widget for my Claude Code sessions because I kept forgetting them
- π‘Australian Payments Plus moves faster with ChatGPT and Codex
- π’Liquid AI - Antidoom (the doom loop remover)
- π’I built a tiny proxy that gives GLM 5.2 vision (or any text LLM) β MIT
2026-07-06
- π‘drinks-sommelier β I created an open-source skill that turns any AI agent into a personal sommelier
- π‘New model: GigaChat3.5-432B-A28B (with day-0 GGUF support!)
- π‘TESLA CAPS EMPLOYEE AI SPENDING AT $200/WEEK EXCEPT FOR GROK (6 MINUTE READ)
- π‘VIBE-CODING PLATFORM BASE44 LAUNCHES OWN MODEL AS AI STARTUPS SEEK DEFENSIBILITY (3 MINUTE READ)
- π‘Meta teases βWatermelonβ model on par with GPT-5.5
- π’Ant Group released LingBot-Vision: DINO-family vision backbones in 4 sizes, and the 0.3B ViT-L matches DINOv3-7B on NYUv2 depth with ~23x fewer params
2026-07-05
- π΄Is the current Open Weight LLM model viable in the long term?
- π΄Looking for an alternative to Antigravity IDE for modular code (~$20)
- π‘Any word on Qwen 3.7 9B? (Also looking for 9B-class alternatives to Qwen 3.5)
- π‘GOOGLE RELEASES NANO BANANA 2 LITE, ITS FASTEST AND CHEAPEST AI IMAGE GENERATOR YET (3 MINUTE READ)
- π‘AMAZON LAUNCHES NEW $1B FDE ORG, FOLLOWING OPENAI AND ANTHROPIC (4 MINUTE READ)
- π‘SEC-GEMINI (GITHUB REPO)
- π‘Claude Fable 5 - Anthropicβs newly reinstated frontier, Mythos-class model
- π’Using llama.cpp with pi
- π’Git for agents with ephemeral runtime
- π’[RELEASE] Supra-Router-51M - a tiny prompt routing model/orchestrator
2026-07-04
- π΄GPT-5.5 Codex reasoning-token clustering may be leading to degraded performance
- π‘ANTHROPIC LAUNCHES CLAUDE SONNET 5 AT A STEEP DISCOUNT TO ITS TOP MODEL AS THE COMPANY RACES TOWARD A BLOCKBUSTER IPO (13 MINUTE READ)
- π‘ANTHROPIC TO RESTORE CLAUDE FABLE ACCESS ON WEDNESDAY (1 MINUTE READ)
- π‘THE DEPARTMENT OF COMMERCE HAS LIFTED EXPORT CONTROLS ON CLAUDE FABLE 5 AND MYTHOS 5 (1 MINUTE READ)
- π‘MEITUAN LAUNCHES LONGCAT-2.0 1.6T PARAMETER MODEL ON APIS (2 MINUTE READ)
- π‘Gemini Omni Flash - Google's video model for video generation and editing
- π’Using local models with Hermes vs Claude code
2026-07-03
- π‘CLAUDE CODE TURNED EVERY ENGINEER INTO THREE. NOW COMPANIES NEED MORE PRODUCT THINKERS (7 MINUTE READ)
- π‘UNDERSTANDING YOUR CLAUDE CODE SPEND: WHAT'S ACTUALLY DRIVING THE COST (6 MINUTE READ)
- π‘REPO-JACKING ANTHROPIC'S CLAUDE COMMUNITY PLUGINS (AND THE SHAS THAT SAVED THEM) (7 MINUTE READ)
- π‘GEMINI'S PERSONALIZED AI IMAGE GENERATION IS NOW FREE FOR US USERS (2 MINUTE READ)
- π‘California gov. Gavin Newsom signed a deal with Anthropic, making Claude available to its local agencies and government at half price, the first AI cleared by the state.
- π’New serious vulnerabilities spiked around release of Claude Mythos Preview
- π’gemma4 e2b is really good, what other small models work on crappy computers?
- π’Whats the catch with SwiReasoning?
- π’I built a fully automated AI video generation & Instagram publishing pipeline in n8n using Gemini and Veo 3. Hereβs how it works.
- π’Small Language Model SLM [D]
2026-07-01
- π΄Non Us Ally should be afraid.
- π‘Open Models - June 2026
- π‘@GoogleDeepMind: Weβre shipping 2 major releases: π Nano Banana 2 Lite: our fastest and cheapest Gemini Image model π Gemini Omni Flash: now available via the Gemini API and in @GoogleAIStudio to help developers gene
- π‘@xai: Introducing Voice Agent Builder: a no-code platform to create human-like voice agents with Grok Voice. Available today at $0.05 / min. http://x.ai/voice
- π‘New PyMuPDF release, supports Markdown [N]
- π‘Redeploying Fable 5 Announcements Jun 30, 2026 Fable 5 returns globally July 1. We're also proposing an industry-wide framework for scoring jailbreak severity, together with Amazon, Microsoft, Google, and other Glasswing partners.
- π’End of an Agony. Real production service that uses LLM to earn money my team had made and now we are so happy that it will die. Here are some of my final "experiences".
- π’Looking for real experiences with AI chat agents in the NDIS space
- π’gemma-4-31B on Cerebras is better than ChatGPT voice mode
- π’A system-level approach to prompt injection: separating instruction and data channels in LLM agents [P]
2026-06-30
- π΄Claude Code is steganographically marking requests
- π‘nvidia/Qwen3.6-27B-NVFP4 just dropped
- π‘GOOGLE IS RATIONING GEMINI ACCESS TO META BECAUSE IT CANNOT PROVIDE ENOUGH COMPUTE (4 MINUTE READ)
- π‘US GOVERNMENT ALLOWS ANTHROPIC LIMITED RELEASE OF AI MODEL THAT SPARKED CYBERSECURITY CONCERNS (3 MINUTE READ)
- π‘Elon Musk said Grok 4.5, trained with supplemental Cursor data, is in private beta at SpaceX and Tesla, claiming its performance matches Anthropicβs Claude Opus.
- π‘Google reportedly capped Metaβs Gemini usage as soaring demand for AI compute outpaced available capacity, delaying some of Metaβs internal projects.
- π’Claude Science
2026-06-29
- π΄Qwen 3.6 27B is the sweet spot for local development
- π΄Amodei: "Open Source Models Will Eat Your Children"
- π΄Itβs time, Sam, itβs time.
- π‘βit works in the demo" is the four most expensive words in AI
- π‘TRUMP ADMINISTRATION ASKS OPENAI TO STAGGER AI MODEL RELEASE (5 MINUTE READ)
- π‘CONTROL AN ANDROID PHONE WITH GEMINI 3.5 FLASH COMPUTER USE (9 MINUTE READ)
- π‘LINUX FOUNDATION AND INDUSTRY LEADERS LAUNCH AKRITES TO DEFEND CRITICAL OPEN SOURCE SOFTWARE AGAINST AI-ENABLED CYBER THREATS (17 MINUTE READ)
- π‘INTRODUCING RIPPLING DATA CLOUD: AI-POWERED BI THAT UNDERSTANDS YOUR WORKFORCE (13 MINUTE READ)
- π’I'm trying to implement CALM paper, and I have some questions. [P]
- π’NASA testing local LLM inference for future space missions
- π’Introducing LongCat-2.0 - , a large-scale MoE language model with 1.6 trillion total parameters and ~48 billion activated per token. This was the stealth model that was on Openrouter under the name 'owl-alpha'.
2026-06-28
- π΄GLM 5.2 beats Claude in our benchmarks
- π΄I used Claude Code to get a second opinion on my MRI
- π΄DFlash support merged into llama.cpp
- π‘WHY BIG AI LABS ARE HIRING SO MANY PHILOSOPHERS (8 MINUTE READ)
- π‘DESIGNING WITH AI: WHY CLAUDE DESIGN IS NOT THE FUTURE OF ENTERPRISE DESIGN (10 MINUTE READ)
- π‘ANTHROPIC REIMAGINES CLAUDE IN SLACK AS NOSY, ALWAYS-ON AGENTIC AI COWORKER (6 MINUTE READ)
- π‘SUPERHUMAN BUYS GPTZERO TO ADD AN AI AUTHENTICITY LAYER (3 MINUTE READ)
- π‘DETECTING MISUSE WITH THE CLAUDE COMPLIANCE API: THE THREAT IS IN THE CONTENT (12 MINUTE READ)
- π’A barebones CPU-only inference engine for Qwen 3, written from scratch in pure C
- π’Show HN: NanoEuler β GPT-2 scale model in pure C/CUDA from scratch
- π’How many of you do use Q1 or Q2 of Big models(100-250B)? How's it?
2026-06-27
- π΄Looking to become proficient in AI for real-world business applications. Where should I start?
- π‘Mythos was the first, now GPT-5.6
- π‘ANTHROPIC WANTS CLAUDE TO BE YOUR NEW SLACK COWORKER (2 MINUTE READ)
- π‘ANALYZING CLAUDE CODE USAGE WITH CLOUDWATCH AND OPENTELEMETRY (6 MINUTE READ)
- π‘GETTY IMAGES ACCUSED AI OF WHOLESALE THEFT. IT'S NOW AN OFFICIAL CHATGPT IMAGE PARTNER (2 MINUTE READ)
- π‘HIGGSFIELD LAUNCHES ENTERPRISE MARKETING AGENTS BUILT ON NVIDIA (3 MINUTE READ)
- π’Ornith 35B is great so far
- π’We built a calibration-aware Q4_K_M quant of Qwen3.5 0.8B that recovers 96.5% of the BF16 gap vs pure llama.cpp Q4_K_M (SpectralQuant)
- π’I built a tool to turn your Claude Code sessions into fine-tuning data for local models
2026-06-26
- π΄US Govt to individually approve who gets GPT 5.6.
- π΄Previewing GPTβ5.6 Sol: a next-generation model
- π‘I'M THE AGENT FOR CLAUDE NOW (5 MINUTE READ)
- π‘OPENAI LAUNCHES DAYBREAK TO AUTOMATE VULNERABILITY PATCHING (4 MINUTE READ)
- π‘A PUBLIC SENTRY KEY IS ALL IT TAKES TO HIJACK CLAUDE CODE, CURSOR, AND CODEX (4 MINUTE READ)
- π‘OPENAI LAUNCHES NEW SECURITY TOOLS AND UPDATES GPT-5.5-CYBER (2 MINUTE READ)
- π‘The Five Eyes cyber agencies released a warning that AI is changing cyber risk in "months, not years," urging execs to harden defenses as attacks speed up.
- π’Show HN: Smart model routing directly in Claude, Codex and Cursor
- π’vulkan: make TP viable by pwilkin Β· Pull Request #25051 Β· ggml-org/llama.cpp
- π’How are you enforcing what AI coding agents can actually do β not just guiding them?
- π’Live Continual Learning in Machine Learning [D]
2026-06-25
- π‘New sampler + verifier *drastically* improves tiny 0.5b model coding performance
- π‘Introducing computer use in Gemini 3.5 Flash
- π‘@GoogleDeepMind: Gemini 3.5 Flash now supports native computer use. This built-in tool lets developers build custom agents that can see and take action across browser, mobile, and desktop interfaces. Find out more β h
- π‘rtx 6000 pro owners, do you regret?
- π’Tensor Split Fix for intel GPU's llama.cpp release b9788
- π’audio.cpp: 12 audio models (Qwen3-TTS, PocketTTS, VeVo2 etc) in 1 C++/ggml runtime β TTS up to 5x faster than Python on CUDA
- π’Prices of graphic cards are going crazy, should I buy a second card though?
2026-06-24
- π΄Qwen-AgentWorld-35B-A3B: a 3B-active MoE trained to simulate MCP, terminal, SWE, Android, web and OS environments
- π΄How do you track per-session costs across STT/LLM/TTS for voice agents?
- π΄The Bank of Korea just released a report about AI productivity
- π‘We go deep onOmnigent: Databricksβ open-source meta-harness for combining, controlling, andsharing agents across Claude Code, Codex, Cursor, Pi, custom agents, and internal tools. Matei explains why c
- π‘Computer use in Gemini 3.5 Flash
- π‘@OpenAI: We have a new version of GPT-5.5 Instant for you, and it's much more fun to talk to. Our most-used model is now better at understanding the intent behind a question and adapting its response according
- π‘@OpenAI: Weβve designed and built our first AI chip: JalapeΓ±o. Designed from the ground up by OpenAI and brought to production with @Broadcom, JalapeΓ±o is purpose-built for the LLM workloads powering ChatGPT,
- π‘@xai: Use the official @MongoDB plugin in Grok Build to query data, optimize indexes, and manage databases.
- π’Big AI labs are hiring philosophers
- π’How Baidu's newly released Unlimited-OCR transcribes dozens of pages in one forward pass
- π’Could it be that there arenβt really any medical LLM APIs available right now? [D]
- π’Do cloud chatbot's system prompts make them stupider?
2026-06-23
- π΄Krea 2 released on Hugging Face
- π΄V100 4-card AI large model, Tesla 128G server
- π‘INTRODUCING THE CLOUDFLARE ONE STACK: AGENT-POWERED DEPLOYMENT (4 MINUTE READ)
- π‘His move lands just days after Gemini co-lead Noam Shazeer left for OpenAI, a major one-two punch of AI talent leaving the company.
- π‘Reflection launched in October to build open frontier systems for government and enterprise, though it has yet to release a public model.
- π‘@MistralAI: Introducing Mistral OCR 4. It creates structure with bounding boxes, block classification, and inline confidence scores in 170 languages. π§΅π
- π‘How GPT-5 helped immunologist Derya Unutmaz solve a 3-year-old mystery
- π’I mapped the KLD of KV cache quantization for Qwen3.6-35B-A3B and Gemma4-E2B QAT
- π’ByteDance Seedance 2.5 is expected in early July, with up to 30s video generation
2026-06-22
- π΄The text in Claude Codeβs βExtended Thinkingβ output
- π‘Claude Code - New artifact integrations for previewing work as live, interactive, sharable pages
- π‘The Outreach System My Friend Used to Generate $235K for His Web Agency
- π‘Do you think dedicated hardware for running local LLMs will become affordable anytime soon?
- π‘Daybreak: Tools for securing every organization in the world
- π‘@OpenAI: Weβre expanding OpenAI Daybreak to help democratize patching vulnerable software at machine speed: - Codex Security plugin: find, validate, and fix vulnerabilities right inside Codex - The full versio
- π’Top-N-Sigma: Remove unconditional softmax+sort by TimNN Β· Pull Request #22645 Β· ggml-org/llama.cpp
- π’TMax: A Simple Recipe for Terminal Agents
- π’GLM-5.2 vs Claude Opus
2026-06-21
- π΄I vibe code apps for a living. Here are my three tips.
- π΄Identity verification on Claude
- π΄Qwen is never going to open source Qwen 3.7, aren't they?
- π΄Predicting model behavior before release by simulating deployment
- π‘8-16 MI50s Minimax M3 @19 tps TG (peak)
- π‘The AI scaling debate always focuses on the question of βhow do we get more GPUs?β but the better question may be:how do we make the most of ones we already have.
- π‘ANTHROPIC SHIPS MAJOR CLAUDE DESIGN OVERHAUL WITH DESIGN SYSTEM IMPORTS, CODE ROUND-TRIPS, AND A FIX FOR ITS TOKEN-BURNING PROBLEM (13 MINUTE READ)
- π‘VERCEL LAUNCHES ENTERPRISE CONTROLS FOR AGENTIC AI INFRASTRUCTURE (3 MINUTE READ)
- π‘INSIDE CLAUDE MANAGED AGENTS: REVERSE ENGINEERING THE SECURITY BOUNDARIES OF ANTHROPIC'S HOSTED AGENT RUNTIME (12 MINUTE READ)
- π’Show HN: Recall β fully-local project memory for Claude Code
- π’Qwen 3.6 27b Abliterated (apostate)
- π’Claude: Elevated Error Rates for Opus 4.8, Opus 4.7, Opus 4.6, and Sonnet 4.6
- π’I forked ik_llama.cpp and added a "--numa mirror" mode to maximize performance on multi-socket CPU systems. Just sharing and looking for testers!
- π’If youβre using AI agents (Claude / Cursor / Copilot)β¦ Youβre probably missing one critical layer: π a safety + cost firewall
2026-06-20
- π΄z.AI as the number 2 gives praise to the number 1 open source model
- π‘The AI scaling debate always focuses on the question of βhow do we get more GPUs?β but the better question may be:how do we make the most of ones we already have.
- π‘GOOGLE GEMINI CO-LEAD NOAM SHAZEER TO JOIN IPO-BOUND OPENAI (5 MINUTE READ)
- π‘ANTHROPIC SHIPS MAJOR CLAUDE DESIGN OVERHAUL (9 MINUTE READ)
- π‘Save hours on meeting prep with Google Gemini
- π‘Databricks launched a slate of new agentic tools at its Data + AI Summit, including LTAP for running AI apps and analytics, and an AI-run customer data platform called CustomerLake.
- π’Itβs time to decentralize model distribution! Introducing Noema Atlas
- π’CortexPrism β open-source agent harness with 24 LLM providers, 5-tier memory, and code intelligence
- π’You can now convert EXL3 quants on Apple Silicon Mac
- π’Board where every tile is an agent
2026-06-19
- π΄Launch HN: TesterArmy (YC P26) β Agents that test web and mobile apps
- π΄GLM-5.2 is above GPT-5.5 in AA-Briefcase, Artificial Analysis' new agentic knowledge work eval
- π‘@xai: Grok models are now available on Databricks Agent Bricks. Bring SpaceXAI's latest models to your enterprise data to power capable AI agents. https://x.ai/news/grok-databricks
- π‘New usage analytics and updated spend controls for enterprises
- π‘Improving health intelligence in ChatGPT
- π‘@AnthropicAI: New Frontier Red Team blog: Phase 2 of Project Fetch, where we test how well Claude can program a robodog. Opus 4.7, on its own, was ~20x faster than last year's best human team aided by Opus 4.1. (T
- π‘Giving GLM-5.2 a spin locally on CPU only! (poor man's rig for big models)
- π’Updates on North Mini Code: 4 bit quant + Ollama + OpenRouter
- π’[NEW MODEL] SupraLabs just released SupraVL-Nano-900k, a Vision-Language Model built entirely from scratch!
2026-06-18
- π΄The captcha arms race is making autonomous web tasks practically impossible
- π΄I released Inflect-Nano, an ultra-extreme tiny 4.63m parameter TTS model.
- π΄Your most useful AI so far? (can be a tool or an agent)
- π΄A robot is sprinting towards you. Do you want it running on Claude or Grok?
- π΄We need a 80-160B model urgently. The unified memory device market needs more Models.
- π‘@OpenAI: Introducing LifeSciBench, a benchmark for measuring and improving how well AI supports real-world life science research. Developed with 173 scientists from biotechnology and pharmaceutical research,
- π‘llama.cpp now supports model management (downloading etc) via API
- π‘Local Qwen isn't a worse Opus, it's a different tool
- π‘Lin Junyang AI Lab Closes Round at $2B Valuation
- π’Quick thoughts on GLM-5.2 (Bonus: Censorship question answers)
2026-06-17
- π΄Donate your coding sessions to an open CC-BY-4.0 dataset to help train open-weight and open source models
- π΄[ECCV 2026] Final Decisions [D]
- π΄Mistral - New family of open-weight models @ July
- π‘GLM-5.2 is now 1st on Design Arena β ahead of the now unavailable Claude Fable 5.
- π‘The agent conversation everyone's skipping: sprawl, shadow agents, and the bill that's coming
- π‘GPTβNL: a sovereign language model for the Netherlands
- π‘@xai: Grok Imagine Video 1.5 is here Our new image-to-video model with sharper realism, better physics and faster generations 𧡠http://grok.com/imagine
- π‘GLM 5.2 API is live, weights are on HF, and ollama has it already
- π’Sharing my DIY AI Memory Framework: Giving LLMs human-like memory (and slashing token costs by 90%)
- π’Looking for an open-source/free alternative to Eigent for Word document template migration (Gemini API)
- π’It looks like Rio 3.5 397B could've simply been a semi-failed embezzling of funding
- π’Cheapest way to run GLM 5.x locally that's not a unified memory system?
- π’Qwen-Robot Suite: A Foundation Model Suite for Physical World Intelligence
2026-06-16
- π΄Stop using Ollama
- π΄Every Al startup is building the same fancy house. On stilts
- π΄Why there is a lack of new 100B-120B models?
- π‘@xai: You can now use your SuperGrok or X Premium subscription inside @warpdotdev. Try it out from Warp Agent Settings and switch to the Grok Build model. https://x.ai/news/grok-warp
- π’Claude Corps
- π’quicktok: a faster tokenizer (exact and byte-identical with tiktoken) [P]
- π’We shipped a customer support agent and our "testing" was basically vibes. Here's what changed after the first real incident.
- π’vLLM has a new streaming parser for Qwen3+ available in nightly
- π’Nex-N2 Pro is the real deal
2026-06-15
- π΄Introducing the Heretic Grimoire: The takedown-resilient, local-first backup system that keeps uncensored models available forever
- π΄Xiaomi is now serving MiMo V2.5 at 1000-3000tps using DFlash & Persistent kernel. DFLash model is out, open-source release promised coming soon
- π‘Claude code + web search API provides excellent leads based on signals from reddit , Linkedin , G2 etc platforms
- π‘EAGLE support merged into llama.cpp
- π‘Command A Plus GGUFs posted
- π‘Introducing the OpenAI Partner Network
- π’An agent that plans with a frontier model but runs most of tokens locally (built it for my own dual-3090 rig)
- π’UI/svg block rendering by ServeurpersoCom Β· Pull Request #24080 Β· ggml-org/llama.cpp
2026-06-14
- π΄This is coming to Chinese open source models pretty soon. - prepare yourself.
- π‘RTX 5080 and RTX 3090 Setup: 80 Tok/s on Qwen 3.6 27B Q8
- π’Codebase getting larger - Qwen3.6-27B starting to compound issues - how to work smartly with this model?
- π’Can we stop dunking on DiffusionGemma and hack it instead?
- π’Making Claude a Chemist
2026-06-13
- π΄We should heavily discourage and moderate cloud API (deepseek api, GLM api, etc.) topics and discussion. This is LOCAL first.
- π‘We should set up a torrent network for open source models.
- π‘@AnthropicAI: The US government, citing national security authorities, has issued an export control directive to suspend all access to Fable 5 and Mythos 5 by any foreign national, whether inside or outside the Uni
- π‘@GoogleDeepMind: Our Robotics Accelerator has launched with 15 startups helping shape the future of physical AI in Europe. π€ This three-month program will connect them with access to our AI stack, Gemini Robotics mod
- π‘@simonw: I got fed up of waiting for OpenAI to bring their much improved gpt-realtime-2 voice conversation model to the ChatGPT product, so I upgraded my OpenAI-WebRTC playground tool to use it and to let you
- π’New model on huggingface
- π’when fable gets banned but it's ok because you've about to download qwen3.7_67b_21a_mythos_father_fable_mother_distilled_ablated_ablitereted_uncensored_agi_sparse_attention_MTP_SuperHOT_q6_maybe_q7_AGI_FINAL.gguf from huggingface
- π’Launch HN: BitBoard (YC P25) β Analytics Workspace for Agents
- π’ZONOS2: real-time TTS with 8B params, 900M active, and high-fidelity voice cloning
2026-06-12
- π΄Qwen Who? DiffusionGemma running at 1,500 tk/s on a Digital Pregnancy Test.
- π΄Gemma 4 Quadruple Release, 12B, 12B QAT, 26B-A4B QAT and 31B QAT Uncensored Heretics!
- π΄Claude Fable is relentlessly proactive
- π‘I distilled my 12 year experience as a product manager and built a free skill that takes you from "I have an app idea" to a real plan and solid MVP
- π‘EAGLE3 has landed in llama.cpp
- π‘is Gemini your main AI model today, or just a secondary option
- π‘PSA: Test your "threads" argument in llama.cpp (+80% performance in my case)
- π‘Anthropic apologizes for invisible Claude Fable guardrails
- π’Are AI agents making traditional software interfaces obsolete?
- π’Claude Fable 5: mid-tier results on coding tasks
- π’πPP-OCRv6 is officially released !
- π’Has anyone noticed that the behavior of the Kimi model has changed?
2026-06-11
- π΄Anthropic's new model Fable will silently handicap work on LLMs [D]
- π‘PRC-linked influence operations are targeting AI debates in the US
- π‘@xai: Grok Voice offers state-of-the-art performance with human-like timing, tone, and warmth. And it's a fraction the price of competitors. Check it out: http://x.ai/api/voice
- π‘@xai: Read more about how Tori, eToro's agent, leverages models and real-time data from SpaceXAI to help consumers analyze market sentiment https://x.ai/news/grok-etoro
- π‘fableExpectations
- π’AMD touts the unified memory architecture
- π’ICMI 2026 Reviews [D]
- π’"system: your previous response was truncated by the output length limit" Help please
- π’qwen3.6-27b tools call loop
- π’Minimax M3 open weights release planned for Friday
2026-06-10
- π΄Claude Fable 5
- π΄If Claude Fable stops helping you, you'll never know
- π‘Without open source LLMs, US AI companies could have already monopoled the technology
- π‘GuideAnts Open Source AI and agents platform
- π‘From data to decisions: how LSEG is scaling trusted AI
- π‘How engineers at Nextdoor use Codex to build without limits
- π‘Fluid, natural voice translation with Gemini 3.5 Live Translate
- π’I'm brand new to running LLMs and the sheer number of tools is overwhelming
- π’Releasing Apodex-1.0 Smol Models (0.8B, 2B, 4B Open-Weights) optimized for Agentic Verification + AgentHarness Evals
- π’Local LLms releases
- π’Can you really replace paid models with a local model?
2026-06-09
- π΄Apple reveals new AI architecture built around Google Gemini models
- π΄Me: Arguing with an AI bot who just posted something on this sub about Llama 3.1.
- π΄Luce Spark: a 35B MoE on a 16 GB GPU, without the offload tax
- π‘Quick note on the QAT of recent
- π‘2X tk/s (from 19.4 -> 38.1 tk/s on 1 x MI50) Playing with a hypothesis like speculative decoding.. but instead of an additional side model, exploiting that I can run multiple computations side-by-side AS IF I had Qwen3.6-27B loaded twice in memory - small quants don't use all the available compute.
- π‘Be very FR - would someone (or you) pay a virtual AI assitant? should I build this? It can read emails, reschedule your whole week, manage your calls, and improve as it is used more.
- π’ggml-webgpu: Improve prefill speeds for k-quants + refactor matmul for Q4/Q5/Q8 and k-quants by yomaytk Β· Pull Request #24225 Β· ggml-org/llama.cpp
- π’Best agentic workflows for finance research, prospecting, and team productivity?
2026-06-08
- π΄llama.cpp Gemma4 MTP support merged!
- π‘Gemma4_31b_fp8 keeping up with Sonnet_4.6_medium in my harness.
- π‘DeepSeek V4 Pro beats GPT-5.5 Pro on precision
- π’QAT variant of Gemma4 26B A4B is not working well for me
- π’Your Agent Is Having an Existential Crisis
- π’We integrated Arc Gate MCP into a Heym enterprise agent β hereβs what tool result poisoning actually looks like in production
2026-06-07
- π΄Cohere's unreleased coding model (early access for localllama)
- π‘Building a Claude-certified developer network: looking for builders to join (free certification path)
- π‘I design with Claude more than Figma now
- π‘Z.ai, we need Air! GLM GGUF wen?
- π’5 Months Later: open-deepthink Now Has Full Knowledge Distillation Mode
- π’I can't wait for all the x250 sample distills of Mythos and GPT-5.6
- π’Research collection of Arxiv whitepapers [R]
- π’Clustering 3x Jetson Nano Orin Supers
2026-06-06
- π΄Donβt act like yβall ainβt thinking it. Iβm just saying the quiet part out loud. /s
- π΄Did Claude increase bugs in rsync?
- π‘PSA: Gemma 4 12B is NOT completely broken for coding and tool calling, you need a special chat template
- π‘@AnthropicAI: New Anthropic Science Blog: Making Claude a chemist. To manipulate a molecule, chemists first need to understand its structure. Their main tool is NMR spectroscopy. We found Opus 4.7 matchesβand on
- π’Qwen3.6-35B-A3B-Uncensored-Claude-4.6-Genesis-APEX-GGUF
2026-06-04
- π΄Introducing Gemma 4 12B: a unified, encoder-free multimodal model
- π΄Here is this month's experimentation: Grocery Agent
- π‘@OpenAI: Weβre bringing new capabilities to GPT-Rosalind, a model series purpose-built for life sciences research at enterprise scale. It brings GPT-5.5βs agentic coding and tool use together with stronger in
- π‘How Endava is redesigning software delivery around AI agents
- π‘Introducing new capabilities to GPT-Rosalind
- π‘How Wasmer used Codex to build a Node.js runtime for the edge
- π‘@xai: Try Grok models on @Cloudflare's AI Gateway!
- π’Trump signs narrower executive order on AI oversight after industry objections
- π’The ways we contain Claude across products
- π’Best Visual Reasoning Model in 2026 (Including APIs) [D]
- π’Gemma 4 QAT confirmed to release soon!
2026-06-03
- π΄Most of the software you rely on was hacked together fast
- π‘Jun 3, 2026 Policy What we learned mapping a yearβs worth of AI-enabled cyber threats
- π‘Jun 2, 2026 Announcements Expanding Project Glasswing
- π‘Jun 1, 2026 Announcements Anthropic confidentially submits draft S-1 to the SEC
- π‘May 28, 2026 Announcements Anthropic raises $65B in Series H funding at $965B post-money valuation
- π‘May 27, 2026 Announcements Anthropic opens Milan office to support Italian enterprise, research, and developers
- π’Another shout out to llama.cpp build b9455 2x3090
- π’I open-sourced a multi-tenant agent memory framework β zero tokens, shared namespaces, self-improving loops
- π’What memory system are you using for your agents?
- π’Holo3.1 35B/9B/4B/0.8B (Qwen 3.5 finetunes)
- π’Mellum & Granite Embedding models are ready on llama.cpp
2026-06-02
- π΄Browse CVPR 2026 papers on PapersWithCode [P]
- π‘@xai: Composer 2.5 is now available inside Grok Build. Composer 2.5 is a fast, highly intelligent model that excels on long-running tasks and following complex instructions.
- π’I asked each of my AI agents to describe their own role. The answers were surprisingly honest.
2026-06-01
- π΄(YT) PewDiePie released his harness/webui
- π΄ChatGPT for Google Sheets exfiltrates workbooks
- π΄Your PDF Is Costing You 3Γ the Tokens and Here's how you can reduce it using markitdown.
- π΄God dammit Qwen
- π‘I ported NVIDIA Parakeet (speech-to-text) to ggml: same output as NeMo, faster, GGUF-quantized, no Python
- π‘@swyx: just a small zoom out on the vibe shift: in Feb 2025 @soumithchintala was talking about his dream of personal, local, private agents, most people didn't believe him. it's June 2026 and @pewdiepie ha
- π‘next MiniMax will be released in ~10 Days
- π’I was a Data Scientist for 10 years before becoming a quadriplegic. For the past 3 months, I built VibeETL from scratch: A lightning-fast, visual Alteryx alternative powered by Polars & React Flow.
2026-05-31
- π΄Anyone else tired of the βClaude just dropped OPUS 4.8 weβre all cookedβ posts on LinkedIn?
- π΄nvidia/Qwen3.6-35B-A3B-NVFP4 Β· Hugging Face
- π‘mudler/Qwen3.6-35B-A3B-Claude-4.7-Opus-Reasoning-Distilled-APEX-MTP-GGUF just released !
- π’Built a minimalist coding agent optimized for memory footprint
- π’Best small model right now (~4B params) that is good with agentic tasks for personal assistant?
- π’Made a program using LocalLLM based on llama.cpp for fellow Book Lovers!
- π’Speed difference between Windows 11 and Linux with llama.cpp: a myth when using medium and large MoE models
2026-05-30
- π΄Notes from the Mistral AI Now Summit
- π‘@xai: grok-build-0.1 is now available via the xAI API in public beta. This is the same model that powers the Grok Build CLI and excels at agentic coding. Priced at $1/m input and $2/m output, itβs extreme
- π‘How Braintrust turns customer requests into code with Codex
- π‘@OpenAI: Windows users, this oneβs for you. Computer use now works on Windows, so Codex can take action on your Windows computer. And with Windows support for Codex in the ChatGPT mobile app, you can start,
- π‘llama : website + unified `llama` binary Β· ggml-org/llama.cpp Β· Discussion #23875
- π’MINISFORUM UM790 Pro
2026-05-29
- π΄Claude Opus 4.8
- π‘@AnthropicAI: We've raised $65 billion in Series H funding at a $965 billion post-money valuation, led by @AltimeterCap, Dragoneer, @Greenoaks, and @sequoia. This investment will help us advance our research and e
- π‘@xai: Grok Build 0.2.7 is now out, with /usage, /login, shared terminals across subagents, and improved image understanding See all updates at https://x.ai/build/changelog
- π‘@MistralAI: We're taking on the hardest problems in the real world ποΈπ π«βοΈ Today at The AI Now Summit, held at the Louvre, we announced AI solutions for aerospace, automotive, energy, and physics. Deployed in p
- π‘Wall-OSS-0.5: 4B VLA with open training code and zero-shot real-robot evaluation[D]
- π‘Claude Code β Everything You Can Configure That the Docs Don't Tell You
- π’llama.cpp B9387 Significant AMD/ROCm PP Update
- π’Researchers let AI models run a simulated society. Claude was the safestβand Grok committed 180 crimes and went extinct within 4 days
- π’Oculus Founders' AI Startup Sesame Launches Human-Like Voice AI App on iOS
- π’Qwen 3.6 27B overdoing it
2026-05-28
- π΄Qwen3.6 huge quality gain from Q4 to Q6 for coding agent
- π‘I built a 103B-token Usenet corpus (1980β2013) β pre-web, human-only, zero AI contamination. Got strong traction on r/ML, thought this community would find it useful.
- π‘CrankGPT by Squeez Labs - hand-cranked edge AI - talk about local AI!!!
- π‘@xai: Use your SuperGrok or X Premium+ subscription in @kilocode. Try grok-build-0.1 for high speed and agentic coding intelligence, available in the Kilo IDE extensions or CLI. https://x.ai/news/grok-ki
- π‘Nvidia LocateAnything - Fast and High-Quality Vision-Language Grounding with Parallel Box Decoding. (10x faster than Qwen3-VL)
- π’i built pengepul: pool multiple claude/gpt accounts behind one local api
- π’Training GPT-like model on non-language series [R]
2026-05-27
- π΄A rare look inside Qwen 3.7βs open source model release approval process:
- π΄Okay 27B made me a believer
- π΄What have you built with Claude Code just for yourself?
- π‘$400 Qwen 3.6-27B Setup - Dual RTX 3060 - 30-50 t/s
- π‘@AnthropicAI: New on the Engineering Blog: The access and permissions we grant agents should evolve with their capabilities. In our own products, we set these parameters through sandboxing, which limits the scope o
- π‘@GoogleDeepMind: Our Gemini for Science tools could help scientists unlock their next breakthrough. π§¬
- π‘@xai: Thank you so much for all the feedback on the Grok Build Beta. Some of you reported hitting limits quickly. Our team found areas to improve caching, so we've reset Grok Build usage limits for all acc
- π’Cactus Hybrid Router: Gemma4-2B can match Gemini-3.1-Flash-Lite by routing 15-55% of tasks to Gemini And Running The Rest Locally.
- π’Folks running qwen 3.6 27b for agentic work. Do you dare to use q4_k_m?
- π’How I gave my AI agent real-time Reddit awareness in ~20 lines of Python using MCP
- π’AI agent for WhatsApp & Telegram
- π’Multi-Agent Parallelism and State Sovereignty in Agent Fleets
2026-05-26
- π΄The Financial Times has published an article about Heretic
- π΄Update on 12x32gb sxm v100 cluster / local AI for legal drafting
- π΄Is there a way to use multiple AI models without paying for 11 different monthly subscriptions?
- π΄Is Qwen3.6 current king for local agentic use?
- π‘@xai: Grok Build is now available in Beta for all SuperGrok and X Premium+ users. Use Plan Mode, create images and videos with Imagine, and build automations or orchestrators with the CLI. Visit http://x.
- π’model : add support for talkie-1930-13b by niklassheth Β· Pull Request #22596 Β· ggml-org/llama.cpp
- π’Running on a macbook, and having issues with crashing? Maybe this will help...
2026-05-25
- π΄PapersWithCode new features - week 1 [P]
- π΄Qwen3.6-35B-A3B vs Gemma4-26B-A4B
- π΄1000 tps generation on Qwen3.6 27B with V100s
- π‘Chromeflow: agentic browser MCP that drives real Chrome with sessions intact (500+ hours hardening)
- π‘hipEngine: Fast Native Qwen 3.6 Inference for RDNA3 (Strix Halo, 7900 XTX)
- π‘What frontend do you guys use?
- π‘I built a mini business consultant using Claude and Lovable as a weekend project,
- π‘MiMo-V2.5-coder
- π’I will not promote: After Two Months of RELAUNCHING I HAVE HIT $85 MRR
- π’I built an AI coding agent that integrates more than 22 AI providers into a single, native plug-and-play environment
2026-05-24
- π΄GPT 5.5 "secret sauce" is just having the thinking be some stupid caveman mode?
- π΄llama.cpp server have built-in native tools (exec_shell, edit_file, etc.)
- π΄Is there any reason for an uncensored model if you have no interest in roleplaying?
- π‘Run Chromeβs tiny Gemma4 (aka Gemini Nano) directly on PC without GPU
- π‘Qwen3.6-35B-A3B-Uncensored-Genesis-APEX-MTP
- π‘NVFP4 + MTP - voilΓ on llama.cpp
- π’llampart 1.0.0 - I released a standalone local web UI for llama-server with translations, extended settings and a polished conversation sidebar
2026-05-23
- π΄NuExtract3 released: open-weight 4B VLM for Markdown, OCR and structured extraction (self-hostable) [P]
- π΄BeeLlama v0.2.0 β major DFlash update. Single RTX 3090: Qwen 3.6 27B up to 164 tps (4.40x), Gemma 4 31B up to 177.8 tps (4.93x). Prompt processing speed near baseline.
- π‘@AnthropicAI: Last month we launched Project Glasswing, our collaborative AI cybersecurity initiative. Since then, we and our partners have found more than ten thousand high- or critical-severity vulnerabilities in
- π‘@GoogleDeepMind: SynthID, our imperceptible watermark for AI-generated content, is expanding to more partners. Weβre also adding new ways to find out if content was generated using AI - just ask in the @GeminiApp or
- π’Microsoft starts canceling Claude Code licenses
- π’meituan-longcat/LongCat-Video-Avatar-1.5 Β· Hugging Face
- π’397B competitor that fits in 256 RAM?
- π’Experimental "Preserve Thinking" Jinja Template for Gemma4 31B in llama.cpp
- π’Is personalized AI memory actually a problem worth solving or am I just coping[D]
2026-05-22
- π΄Waiting for Qwen 3.7 open weight... The new King has arrived...
- π΄Qwen3.6 35Ba3 has changed my workflows and even how I use my computer
- π‘LatitudeGames/Equinox-31B Β· Hugging Face
- π‘Launch HN: Runtime (YC P26) β Sandboxed coding agents for everyone on a team
- π‘For everyone that uses OpenCode / Pi - Heres your promptprocessing fix!
- π‘AdventHealth advances whole-person care with OpenAI
- π‘Weβre launching the Google DeepMind Accelerator program in Asia Pacific to tackle environmental risks
- π’Which agent should I use for coding and notetaking?
- π’New Release of ROCm based MLX LLM Engine - lemon-mlx-engine
- π’Low-level coding dataset
- π’ztok β a fast multithreaded tokenizer in Zig that loads tiktoken / HF / SentencePiece and is 2β5Γ faster
2026-05-21
- π΄Qwen will release another 27B with high probability
- π‘Qwen 3.6 35B GGUF: NTP vs MTP quantization results across GPUs and CPUs
- π‘@GoogleDeepMind: How can you accelerate your day to day research workflow? By giving AI the right scientific toolkit. We launched Science Skills for Google @Antigravity, integrating insights from over 30 major life
- π‘@GoogleDeepMind: Gemini 3.5 Flash has landed.
- π‘Same task in github-copilot, pi, claude-code, and opencode with Qwen3.6 27B
- π’How can you stop your model from looping
2026-05-20
- π΄Gemini 3.5 Flash
- π΄Do you guys actually think AI agents can replace people for bigger tasks anytime soon?
- π‘LM Studio finally added support for MTP Speculative Decoding
- π‘Co-Scientist (Nature 2026-05-19): 5+1 Gemini agents, tournament-of-ideas, prod
- π‘@OpenAI: People are generating over 1.5 billion images a week in ChatGPT. Researcher @kenjihata joins Product lead @adele__li and host @AndrewMayne to explore the new use cases and trends emerging since the l
- π‘@OpenAI: Introducing OpenAI Guaranteed Capacity: a new offering that enables customers to guarantee long-term access to OpenAI compute. Weβve made long-term investments in infrastructure, partnerships, and ca
- π‘Carbon: Decoding the Language of Life
- π’Qwen3.7 Max scored by Artificial Analysis, 27B/35B waiting room
- π’Gemini CLI will stop working from June 18, 2026
- π’Letβs talk quants of Gemma and Qwen - 16 vs Q8 vs Q4 - any experiences?
- π’Gemma 4 MTP with LlamaCPP
2026-05-19
- π΄Qwen cant wait to release 3.7 models
- π΄Reviving PapersWithCode (by Hugging Face) [P]
- π΄Qwen is cooking hard
- π‘What happens to local LLM if/when LLMs are no longer released for free?
- π‘Released a free 9.8M doc Indic multilingual corpus β Hindi, Bengali, Tamil, Telugu + 7 more (CC0, HuggingFace) [P]
- π’MTP (Multi-Token Prediction): 2x Faster Token Generation on AMD Strix Halo & Radeon 9700 AI Pro
- π’favorite Agentic Coding Harness
- π’We built a tool that installs frameworks like ComfyUI, Ollama, OpenWebUI etc on any cloud GPU in one command and saves your whole setup between sessions [R]
2026-05-18
- π΄"Generate a photorealistic realtime render of a human face with webGL" (Qwen3.5-122B-A10B UD-Q3_K_XL)
- π‘llama: avoid copying logits during prompt decode in MTP by am17an Β· Pull Request #23198 Β· ggml-org/llama.cpp
- π‘The power of structured workflows and small local models
- π’could refusal layers be masking dialect-conditioned safety failures in MoE models [d]
2026-05-17
- π΄MTP PR Merged!!!
- π΄That's a good news...
- π΄AI Agents are pushing us towards Transactional Development β Spin Up, Build, Ship, Discard.
- π΄Anthropic and OpenAI claims that their models are so powerful that it can βbreakβ their boxβ¦but what so special about their agent implementation?
- π΄MTP support merged into llama.cpp
- π‘some things i learned the hard way using claude design
- π‘G4-Meromero-31B-Uncensored-Heretic Is Out Now, a Finetune of Gemma 4 31B It Designed for Creative Tasks, With Kld of 0.0100 and 15/100 Refusals!
- π‘@xai: You can now use X Premium subscriptions in Hermes Agent, and Hermes Agent can now search X posts. https://x.ai/news/grok-hermes
- π’webui: support video files as input by foldl Β· Pull Request #22830 Β· ggml-org/llama.cpp
- π’OpenAI and Government of Malta partner to roll out ChatGPT Plus to all citizens
- π’Looking to migrate off of Ollama and LMStudio
2026-05-16
- π΄Opencode you naughty minx
- π΄I've shipped 3 products this year. None of them have users. Here's my problem.
- π΄Dynamically allocating compute budget to hard set of problems and evolving the sections with Qwen-35B-A3B gets you near GPT-5.4-xHigh on HLE
- π‘Orthrus-Qwen3-8B : up to 7.8Γtokens/forward on Qwen3-8B, frozen backbone, provably identical output distribution
- π‘KDD 2026 Cycle 2 Results [D]
- π‘[FOUNDING] SupraLabs - real open-source AI models for you!
- π‘@xai: You can now use your @grok subscription inside @NousResearch Hermes Agent. http://x.ai/news/grok-hermes
- π’Luce Megakernal: Why nobody is taking about this?
- π’[R] Which LLMs are actually best for bleeding-edge Linux/ML debugging workflows in 2026? [R]
2026-05-15
- π΄The RTX 5000 PRO (48GB) arrived and it is better than I expected.
- π΄Codex is now in the ChatGPT mobile app
- π΄How Claude Code works in large codebases
- π‘China modded GPU (eg. 4090 48gb) --> I'm gonna figure it out. IS THERE NO ONE ELSE CURIOUS??
- π‘@xai: An early beta of Grok Build, an agentic CLI for coding, building apps, and automating workflows is now available for SuperGrok Heavy subscribers. Through this early beta, we will improve the model an
- π‘I Let a Small Model Train on Its Own Mistakes. It Reached 80% on HumanEval and Beat GPT-3.5 on Math
- π‘Claude for Legal
- π‘@AnthropicAI: Weβre partnering with the Gates Foundation, committing $200 million in grants, Claude credits, and technical support to programs in global health, life sciences, education, agriculture, and economic m
- π’RDNA3 Flash Attention fix just dropped by llama.cpp b9158
- π’Show HN: GlycemicGPT β Open-source AI-powered diabetes management
- π’RelaxAI β UK sovereign LLM inference at 80% cheaper than OpenAI/Claude
- π’I have (even faster) DeepSeek V4 Pro at home
- π’I got tired of OpenClaw skills having no actual usage so I spent 3 weeks building one.
2026-05-14
- π΄TextGen is now a native desktop app. Open-source alternative to LM Studio (formerly text-generation-webui).
- π΄DELIGHT β self-hosted AI engineering autopilot: local LLM + browser farm + repo graph + P2P compute
- π΄Claude for Small Business
- π’What revenue model would you guys suggest for our automation orchestration platform open to public agents as a marketplace?
- π’A Claude Code and Codex Skill for Deliberate Skill Development
- π’Simpler self hosted alt to Open WebUI
- π’The "the future is fictional" problem of many local LLMs
2026-05-13
- π΄Stop wasting electricity
- π΄TabPFN-3 just released: a pre-trained tabular foundation model for up to 1M rows [R][N]
- π‘Let's build claude code from scratch!
- π‘Is using vLLM actually worth it if you aren't serving the model to other people?
- π‘I think AI businesses are quietly shifting from hype products to operational tools
- π’I've seen a lot of folks ask "can local LLMs actually do anything useful?"
- π’OpenAI API now requires separate pay-as-you-go billing β what alternatives are you all using?
- π’Small local model for questions on German grammar
2026-05-12
- π΄MTP on Unsloth
- π‘@OpenAI: Introducing Daybreak: frontier AI for cyber defenders. Daybreak brings together the most capable OpenAI models, Codex, and our security partners to accelerate cyber defense and continuously secure so
- π‘Will there be any more Qwen3.6 series models?
- π‘How ChatGPT adoption broadened in early 2026
- π‘@AnthropicAI: New Anthropic research: Teaching Claude why. Last year we reported that, under certain experimental conditions, Claude 4 would blackmail users. Since then, weβve completely eliminated this behavior.
- π‘@OpenAI: Today weβre launching the OpenAI Deployment Company to help businesses build and deploy AI. It's majority-owned and controlled by OpenAI. It brings together 19 leading investment firms, consultancies
- π’I catalogued every way local models break JSON output and built a repair library, here's what I found across 288 model calls
- π’Llama models: still valuable for finetuning or surpassed by everything new?
- π’Claude Platform on AWS
2026-05-11
- π΄I have DeepSeek V4 Pro at home
- π΄How I set up an AI agent to handle invoicing bill pay and expense tracking through my bank via MCP
- π‘Any implementations similar to D4RT? [D]
- π‘Put my 4 years of SEO experience into a claude skill so you don't have to figure it out yourself
- π‘ExLlamaV3 Major Updates!
- π’What's actually moving the needle on agent token bills?
- π’How Fast Does Claude, Acting as a User Space IP Stack, Respond to Pings?
- π’The Qwen 3.6 35B A3B hype is real!!!
- π’Show HN: adamsreview β better multi-agent PR reviews for Claude Code
2026-05-10
- π‘NVIDIA AI Releases Star Elastic: One Checkpoint that Contains 30B, 23B, and 12B Reasoning Models with Zero-Shot Slicing
- π‘Pi and Qwen3.6 27B make setting up Archlinux really easy.
- π‘More Qwen3.6-27B MTP success but on dual Mi50s
- π’Building a AI teacher-assistance software, Assistance needed.
- π’Would a community-driven AI agent lab help people actually ship agents?
- π’Exactly a year ago, I started working on an MCP server I launched on reddit that became by far my most active open source project!
- π’Gemini API File Search is now multimodal
- π’I am overwhelmed by Harnesses
2026-05-09
- π΄vLLM ROCm has been added to Lemonade as an experimental backend
- π΄A recent experience with ChatGPT 5.5 Pro
- π΄I'm kinda good at getting users for ai tools through reddit - could I make money?
- π‘Teaching Claude Why
- π‘Reports suggest DeepSeek is seeking $7.35 billion in funding and plans to release its V4.1 update next month.
- π‘@OpenAI: Chain of thought monitors are a key layer of defense against AI agent misalignment. To preserve monitorability, we avoid penalizing misaligned reasoning during RL. We found a limited amount of accide
- π‘Using Claude Code: The unreasonable effectiveness of HTML
- π‘new MoE from ai2, EMO
- π’Formalizing statistical learning theory in Lean 4 [R]
- π’How long for llama.cpp official support of MTP?
- π’FEEDBACK FOR MY APP
2026-05-08
- π‘Agentic workflows
- π‘AlphaEvolve: Gemini-powered coding agent scaling impact across fields
- π‘You can now read Gemma 3's mind
- π‘Scaling Trusted Access for Cyber with GPT-5.5 and GPT-5.5-Cyber
- π‘@AnthropicAI: Weβre donating Petri, our open-source alignment tool, to @meridianlabs_ai, so its development can continue independently. Working with Meridian Labs, weβve also released a major update that improves
- π’Quantization and Fast Inference (MEAP) - How much performance are you actually getting from quantization in production? [D]
- π’Hardening Firefox with Claude Mythos Preview
2026-05-07
- π‘Qwen3.6 27B uncensored heretic v2 Native MTP Preserved is Out Now With KLD 0.0021, 6/100 Refusals and the Full 15 MTPs Preserved and Retained, Available in Safetensors, GGUFs and NVFP4s formats.
- π‘What Anthropic's 'Dreaming' feature release made me notice about my own ClawVault agent's memory
- π‘@xai: SpaceXAI will provide @AnthropicAI with access to Colossus 1, one of the worldβs largest and fastest-deployed AI supercomputers, to provide additional capacity for Claude β http://x.ai/news/anthropic-
- π’why llama.cpp canβt combine speculative decode methods?
- π’Qwen 3.6?
2026-05-06
- π΄Gemma 4 MTP released
- π΄Sr Software Engineer - Haven't written a line of code in months
- π‘MTP on strix halo with llama.cpp (PR #22673)
- π’Tired of copy-pasting prompts between Claude and Codex tabs: built a small file-backed queue that automates the handoff
- π’Bleeding Llama: Critical Unauthenticated Memory Leak in Ollama
- π’Qwen 3.6 27B MTP on v100 32GB: 54 t/s
- π’Solidity LM surpasses Opus
2026-05-05
- π΄White House Considers Vetting A.I. Models Before They Are Released
- π΄Peanut - Text to Image Model (Open Weights coming soon)
- π‘FastDMS: 6.4X KV-cache compression running faster than vLLM BF16/FP8
- π‘Claude Code structure that didnβt break after 2β3 real projects
- π’As MTP prepares to land in llama.cpp, Models that support MTP
2026-05-03
- π΄Qwen3.6-27B vs Coder-Next
- π΄Karpathy's MicroGPT running at 50,000 tps on an FPGA
- π΄GPT 5.5 just leaked its chain of thought to me in codex, and it looks like an idea from 5 months ago in this sub.
- π‘I made a visualizer for Hugging Face models
- π’What a time to be alive from 1tk/sec to 20-100tk/sec for huge models
2026-04-30
- π΄Large Language Models Explore by Latent Distilling
- π‘ClawGym: A Scalable Framework for Building Effective Claw Agents
- π‘Introducing Advanced Account Security
- π‘Enabling a new model for healthcare with AI co-clinician
- π’Operating-Layer Controls for Onchain Language-Model Agents Under Real Capital
2026-04-28
- π‘ReVSI: Rebuilding Visual Spatial Intelligence Evaluation for Accurate Assessment of VLM 3D Reasoning
- π‘Apr 28, 2026 Announcements Claude for Creative Work
- π‘Apr 27, 2026 Announcements Anthropic names Theo Hourmouzis General Manager of Australia & New Zealand and officially opens Sydney office
- π‘Apr 24, 2026 Announcements An update on our election safeguards
- π‘Apr 24, 2026 Announcements Anthropic and NEC collaborate to build Japanβs largest AI engineering workforce
2026-04-27
- π΄DiffNR: Diffusion-Enhanced Neural Representation Optimization for Sparse-View 3D Tomographic Reconstruction
- π‘OpenAI available at FedRAMP Moderate
- π‘Contexts are Never Long Enough: Structured Reasoning for Scalable Question Answering over Long Document Sets
- π’Building a Precise Video Language with Human-AI Oversight
2026-04-22
- π΄Tstars-Tryon 1.0: Robust and Realistic Virtual Try-On for Diverse Fashion Items
- π‘SmartPhotoCrafter: Unified Reasoning, Generation and Optimization for Automatic Photographic Image Editing
- π‘Making ChatGPT better for clinicians
- π‘Workspace agents
- π‘Introducing workspace agents in ChatGPT
- π‘Introducing OpenAI Privacy Filter
2026-04-20
- π΄Maximal Brain Damage Without Data or Optimization: Disrupting Neural Networks via Sign-Bit Flips
- π‘Qwen3.5-Omni Technical Report
- π‘PersonaVLM: Long-Term Personalized Multimodal LLMs
- π‘OpenAI helps Hyatt advance AI among colleagues
- π’Cut Your Losses! Learning to Prune Paths Early for Efficient Parallel Reasoning
2026-04-17
- π΄HY-World 2.0: A Multi-Modal World Model for Reconstructing, Generating, and Simulating 3D Worlds
- π΄How to Fine-Tune a Reasoning Model? A Teacher-Student Cooperation Framework to Synthesize Student-Consistent SFT Data
- π‘Product Apr 17, 2026 Introducing Claude Design by Anthropic Labs Today, weβre launching Claude Design, a new Anthropic Labs product that lets you collaborate with Claude to create polished visual work like designs, prototypes, slides, one-pagers, and more.
- π‘Announcements Feb 4, 2026 Claude is a space to think Weβve made a choice: Claude will remain ad-free. We explain why advertising incentives are incompatible with a genuinely helpful AI assistant, and how we plan to expand access without compromising user trust.
- π‘Dive into Claude Code: The Design Space of Today's and Future AI Agent Systems
2026-04-16
- π΄Seedance 2.0: Advancing Video Generation for World Complexity
- π΄RationalRewards: Reasoning Rewards Scale Visual Generation Both Training and Test Time
- π‘OccuBench: Evaluating AI Agents on Real-World Professional Tasks via Language World Models
- π‘Introducing Claude Opus 4.7 Product Apr 16, 2026 Our latest Opus model brings stronger performance across coding, agents, vision, and multi-step tasks, with greater thoroughness and consistency on the work that matters most.
- π‘Introducing GPT-Rosalind for life sciences research
- π‘Accelerating the cyber defense ecosystem that protects us all
- π’Exploration and Exploitation Errors Are Measurable for Language Model Agents
2026-04-15
- π‘Nemotron 3 Super: Open, Efficient Mixture-of-Experts Hybrid Mamba-Transformer Model for Agentic Reasoning
- π‘Gemini 3.1 Flash TTS: the next generation of expressive AI speech
- π’BERT-as-a-Judge: A Robust Alternative to Lexical Methods for Efficient Reference-Based LLM Evaluation
- π’The Blind Spot of Agent Safety: How Benign User Instructions Expose Critical Vulnerabilities in Computer-Use Agents
2026-04-08
- π΄Video-MME-v2: Towards the Next Stage in Benchmarks for Comprehensive Video Understanding
- π‘Featured An update on recent Claude Code quality reports We traced recent reports of Claude Code quality issues to three separate changes. Here's what happened and what we're changing.
- π‘Scaling Managed Agents: Decoupling the brain from the hands Apr 08, 2026
- π‘Claude Code auto mode: a safer way to skip permissions Mar 25, 2026
- π‘Harness design for long-running application development Mar 24, 2026
- π‘Eval awareness in Claude Opus 4.6βs BrowseComp performance Mar 06, 2026
- π’ThinkTwice: Jointly Optimizing Large Language Models for Reasoning and Self-Refinement
- π’How Well Do Agentic Skills Work in the Wild: Benchmarking LLM Skill Usage in Realistic Settings
2026-04-06
- π΄GrandCode: Achieving Grandmaster Level in Competitive Programming via Agentic Reinforcement Learning
- π‘Agentic-MME: What Agentic Capability Really Brings to Multimodal Intelligence?
- π’Communicating about Space: Language-Mediated Spatial Integration Across Partial Views
- π’AgentHazard: A Benchmark for Evaluating Harmful Behavior in Computer-Use Agents
2026-04-02
- π΄ClawKeeper: Comprehensive Safety Protection for OpenClaw Agents Through Skills, Plugins, and Watchers
- π΄Terminal Agents Suffice for Enterprise Automation
- π‘Embarrassingly Simple Self-Distillation Improves Code Generation
- π‘Codex now offers more flexible pricing for teams
- π’QuitoBench: A High-Quality Open Time Series Forecasting Benchmark
- π’GaussianGPT: Towards Autoregressive 3D Gaussian Scene Generation
2026-04-01
- π΄FIPO: Eliciting Deep Reasoning with Future-KL Influenced Policy Optimization
- π΄CARLA-Air: Fly Drones Inside a CARLA World -- A Unified Infrastructure for Air-Ground Embodied Intelligence
- π‘GEMS: Agent-Native Multimodal Generation with Memory and Skills
- π‘Gradient Labs gives every bank customer an AI account manager
- π‘Project Imaging-X: A Survey of 1000+ Open-Access Medical Imaging Datasets for Foundation Model Development
2026-03-31
- π΄TAPS: Task Aware Proposal Distributions for Speculative Sampling
- π΄Gen-Searcher: Reinforcing Agentic Search for Image Generation
- π‘Accelerating the next phase of AI
- π‘HISA: Efficient Hierarchical Indexing for Fine-Grained Sparse Attention
- π’GEditBench v2: A Human-Aligned Benchmark for General Image Editing
- π’ImagenWorld: Stress-Testing Image Generation Models with Explainable Human Evaluation on Open-ended Real-World Tasks
2026-03-27
- π΄Calibri: Enhancing Diffusion Transformers via Parameter-Efficient Calibration
- π‘Voxtral TTS
- π‘STADLER reshapes knowledge work at a 230-year-old company
- π’MACRO: Advancing Multi-Reference Image Generation with Structured Long-Context Data
- π’AVControl: Efficient Framework for Training Audio-Visual Controls
- π’VFIG: Vectorizing Complex Figures in SVG with Vision-Language Models
2026-03-26
- π΄CUA-Suite: Massive Human-annotated Video Demonstrations for Computer-Use Agents
- π΄Why Does Self-Distillation (Sometimes) Degrade the Reasoning Capability of LLMs?
- π‘T-MAP: Red-Teaming LLM Agents with Trajectory-aware Evolutionary Search
- π‘Gemini 3.1 Flash Live: Making audio AI more natural and reliable
2026-03-20
- π΄SAMA: Factorized Semantic Anchoring and Motion Alignment for Instruction-Guided Video Editing
- π΄Nemotron-Cascade 2: Post-Training LLMs with Cascade RL and Multi-Domain On-Policy Distillation
- π‘FASTER: Rethinking Real-Time Flow VLAs
- π’F2LLM-v2: Inclusive, Performant, and Efficient Embeddings for a Multilingual World
2026-03-17
- π΄AI Can Learn Scientific Taste
- π‘Introducing GPT-5.4 mini and nano
- π‘OpenAI Japan announces Japan Teen Safety Blueprint to put teen safety first
- π‘Equipping workers with insights about compensation
- π‘Measuring progress toward AGI: A cognitive framework
- π‘EnterpriseOps-Gym: Environments and Evaluations for Stateful Agentic Planning and Tool Use in Enterprise Settings
- π’Mixture-of-Depths Attention
- π’Effective Distillation to Hybrid xLSTM Architectures
- π’Safe and Scalable Web Agent Learning via Recreated Websites
2026-03-16
- π΄Multimodal OCR: Parse Anything from Documents
- π‘Cheers: Decoupling Patch Details from Semantic Representations Enables Unified Multimodal Comprehension and Generation
- π‘OmniForcing: Unleashing Real-time Joint Audio-Visual Generation
- π‘Video Streaming Thinking: VideoLLMs Can Watch and Think Simultaneously
- π‘daVinci-Env: Open SWE Environment Synthesis at Scale
- π’Visual-ERM: Reward Modeling for Visual Equivalence
2026-03-13
- π΄Strategic Navigation or Stochastic Search? How Agents and Humans Reason Over Document Collections
- π‘Video-Based Reward Modeling for Computer-Use Agents
- π’DVD: Deterministic Video Depth Estimation with Generative Priors
- π’Trust Your Critic: Robust Reward Modeling and Reinforcement Learning for Faithful Image Editing and Generation
2026-03-11
- π΄Geometry-Guided Reinforcement Learning for Multi-view Consistent 3D Scene Editing
- π‘Designing AI agents to resist prompt injection
- π‘Fish Audio S2 Technical Report
- π’Stepping VLMs onto the Court: Benchmarking Spatial Intelligence in Sports
- π’Reading, Not Thinking: Understanding and Bridging the Modality Gap When Text Becomes Pixels in Multimodal LLMs
- π’Are Audio-Language Models Listening? Audio-Specialist Heads for Adaptive Audio Steering
2026-03-09
- π΄Penguin-VL: Exploring the Efficiency Limits of VLM with LLM-based Vision Encoders
- π‘Progressive Residual Warmup for Language Model Pretraining
- π‘Reasoning Models Struggle to Control their Chains of Thought
- π’Physical Simulator In-the-Loop Video Generation
- π’HiMAP-Travel: Hierarchical Multi-Agent Planning for Long-Horizon Constrained Travel
2026-03-05
- π΄Helios: Real Real-Time Long Video Generation Model
- π΄T2S-Bench & Structure-of-Thought: Benchmarking and Prompting Comprehensive Text-to-Structure Reasoning
- π‘Introducing GPT-5.4
- π‘GPT-5.4 Thinking System Card
- π‘Introducing the Adoption news channel
- π‘Introducing ChatGPT for Excel and new financial data integrations
- π‘VfL Wolfsburg turns ChatGPT into a club-wide capability
- π’Phi-4-reasoning-vision-15B Technical Report
2026-03-03
- π΄SWE-rebench V2: Language-Agnostic SWE Task Collection at Scale
- π‘RubricBench: Aligning Model-Generated Rubrics with Human Standards
- π‘CHIMERA: Compact Synthetic Data for Generalizable LLM Reasoning
- π‘GPT-5.3 Instant System Card
- π‘GPT-5.3 Instant: Smoother, more useful everyday conversations
- π‘Gemini 3.1 Flash-Lite: Built for intelligence at scale
- π’MMR-Life: Piecing Together Real-life Scenes for Multimodal Multi-image Reasoning
2026-02-27
- π΄From Blind Spots to Gains: Diagnostic-Driven Iterative Training for Large Multimodal Models
- π΄MobilityBench: A Benchmark for Evaluating Route-Planning Agents in Real-World Mobility Scenarios
- π‘Introducing the Stateful Runtime Environment for Agents in Amazon Bedrock
- π’AgentDropoutV2: Optimizing Information Flow in Multi-Agent Systems via Test-Time Rectify-or-Reject Pruning
2026-02-26
- π‘DualPath: Breaking the Storage Bandwidth Bottleneck in Agentic LLM Inference
- π‘DreamID-Omni: Unified Framework for Controllable Human-Centric Audio-Video Generation
- π‘OpenAI Codex and Figma launch seamless code-to-design experience
- π’GUI-Libra: Training Native GUI Agents to Reason and Act with Action-aware Supervision and Partially Verifiable RL
2026-02-18
- π‘Introducing OpenAI for India
- π‘Introducing EVMbench
- π‘A new way to express yourself: Gemini can now create music
- π‘A Trajectory-Based Safety Audit of Clawdbot (OpenClaw)
- π’ResearchGym: Evaluating Language Model Agents on Real-World AI Research
- π’HLE-Verified: A Systematic Verification and Structured Revision of Humanity's Last Exam
2026-02-17
- π΄BitDance: Scaling Autoregressive Generative Models with Binary Tokens
- π‘Nanbeige4.1-3B: A Small General Model that Reasons, Aligns, and Acts
- π‘REDSearcher: A Scalable and Cost-Efficient Framework for Long-Horizon Search Agents
- π’Query as Anchor: Scenario-Adaptive User Representation via Large Language Model
- π’Data Darwinism Part I: Unlocking the Value of Scientific Data for Pre-training
2026-02-16
- π΄Less is Enough: Synthesizing Diverse Data in Feature Space of LLMs
- π΄MedXIAOHE: A Comprehensive Recipe for Building Medical MLLMs
- π‘Towards Universal Video MLLMs with Attribute-Structured and Quality-Verified Instructions
- π‘OneVision-Encoder: Codec-Aligned Sparsity as a Foundational Principle for Multimodal Intelligence
- π’GeoAgent: Learning to Geolocate Everywhere with Reinforced Geographic Characteristics
2026-02-13
- π΄DeepGen 1.0: A Lightweight Unified Multimodal Model for Advancing Image Generation and Editing
- π‘Learning beyond Teacher: Generalized On-Policy Distillation with Reward Extrapolation
- π‘GPT-5.2 derives a new result in theoretical physics
- π‘Introducing Lockdown Mode and Elevated Risk labels in ChatGPT
- π‘Scaling social science research
- π’Think Longer to Explore Deeper: Learn to Explore In-Context via Length-Incentivized Reinforcement Learning
2026-02-12
- π΄Step 3.5 Flash: Open Frontier-Level Intelligence with 11B Active Parameters
- π΄GENIUS: Generative Fluid Intelligence Evaluation Suite
- π‘ASA: Training-Free Representation Engineering for Tool-Calling Agents
- π‘Introducing GPT-5.3-Codex-Spark
- π‘Gemini 3 Deep Think: Advancing science, research and engineering
- π‘Towards Autonomous Mathematics Research
- π’TimeChat-Captioner: Scripting Multi-Scene Videos with Time-Aware and Structural Audio-Visual Captions
2026-02-11
- π΄OPUS: Towards Efficient and Principled Data Selection in Large Language Model Pre-training in Every Iteration
- π΄Code2World: A GUI World Model via Renderable Code Generation
- π‘Chain of Mindset: Reasoning with Adaptive Cognitive Modes
- π‘P1-VL: Bridging Visual Perception and Scientific Reasoning in Physics Olympiads
2026-02-09
- π΄F-GRPO: Don't Let Your Policy Learn the Obvious and Forget the Rare
- π‘Baichuan-M3: Modeling Clinical Inquiry for Reliable Medical Decision-Making
- π‘Testing ads in ChatGPT
- π‘Bringing ChatGPT to GenAI.mil
- π‘Accelerating Mathematical and Scientific Discovery with Gemini Deep Think
- π’MSign: An Optimizer Preventing Training Instability in Large Language Models via Stable Rank Restoration
- π’Judging What We Cannot Solve: A Consequence-Based Approach for Oracle-Free Evaluation of Research-Level Math
2026-02-05
- π‘Training Data Efficiency in Multimodal Process Reward Models
- π‘OmniSIFT: Modality-Asymmetric Token Compression for Efficient Omni-modal Large Language Models
- π‘GPT-5 lowers the cost of cell-free protein synthesis
- π‘Introducing Trusted Access for Cyber
- π‘Introducing OpenAI Frontier
- π’EgoActor: Grounding Task Planning into Spatial-aware Egocentric Actions for Humanoid Robots via Visual-Language Models
- π’Rethinking the Trust Region in LLM Reinforcement Learning
2026-02-04
- π΄AOrchestra: Automating Sub-Agent Creation for Agentic Orchestration
- π΄No Global Plan in Chain-of-Thought: Uncover the Latent Planning Horizon of LLMs
- π’SWE-World: Building Software Engineering Agents in Docker-Free Environments
- π’SWE-Master: Unleashing the Potential of Software Engineering Agents via Post-Training
2026-02-03
- π΄Kimi K2.5: Visual Agentic Intelligence
- π΄Vision-DeepResearch: Incentivizing DeepResearch Capability in Multimodal Large Language Models
- π‘Vision-DeepResearch Benchmark: Rethinking Visual and Textual Search for Multimodal Large Language Models
- π‘The Sora feed philosophy
- π’SWE-Universe: Scale Real-World Verifiable Environments to Millions