AI Watchtower Briefing — 2026-07-26

🔴 High Significance

Model Releases

🔴 🤗 zai-org/GLM-5.2 (827,191 downloads) — score 95 Sources: huggingface_models

Author: | Downloads: 827,191 | Likes: 4474

🔴 💬 So GPT 6 isn’t it? — score 93 Sources: reddit/r/OpenAI

🔴 💬 CEO of Hugging Face: "In the spirit of transparency, here’s what I asked OpenAI" — score 89 Sources: reddit/r/LocalLLaMA

clem 🤗 on 𝕏: https://x.com/ClementDelangue/status/2081056675558195657 • Radical transparency: let’s release the traces from the “rogue” agents so the entire research community can study what happened. • More capabilities for defenders: let’

🔴 🤗 baidu/Unlimited-OCR (2,593,460 downloads) — score 85 Sources: huggingface_models

Author: | Downloads: 2,593,460 | Likes: 3199

🔴 💬 Since GPT image 2 performed well. Do you guys think there'll be a GPT Image 2.5 soon? — score 79 Sources: reddit/r/OpenAI

OpenAI released GPT Image 1.5 after the massive success of GPT image 1. Do you guys think there'll be a possibility for GPT Image 2 to be following the same path?

Omitted 5 additional model releases items from the main section; see raw data and source-specific sections below.

Developer Tools

🔴 🐙 CoreBunch/Instatic — The open-source alternative to Webflow, Framer and WordPress. Agentic self-hosted visual CMS outputting clean static pages. Users, roles, plugins, content, database, it's all there. — score 98 Sources: github_trending

The open-source alternative to Webflow, Framer and WordPress. Agentic self-hosted visual CMS outputting clean static pages. Users, roles, plugins, content, database, it's all there.

🔴 🐙 ComposioHQ/awesome-claude-skills — A curated list of awesome Claude Skills, resources, and tools for customizing Claude AI workflows — score 97 Sources: github_trending

A curated list of awesome Claude Skills, resources, and tools for customizing Claude AI workflows

🔴 🐙 anthropics/claude-cookbooks — A collection of notebooks/recipes showcasing some fun and effective ways of using Claude. — score 95 Sources: github_trending

A collection of notebooks/recipes showcasing some fun and effective ways of using Claude.

🔴 💬 I implemented the YOLO26n model inference from scratch using ARM64 Assembly Language (No framework) [P] — score 94 Sources: reddit/r/MachineLearning

This was my Bachelor's Final Project: implementing YOLO26n inference completely from scratch using ARM64 Assembly Language and C, without relying on existing inference frameworks. The goal was to understand how modern neural network inference engines work at a low level and explore optimization tech

🔴 💬 How is your ai agent doing searches and not getting blocked? — score 94 Sources: reddit/r/AIAgents

I mostly use Pi/Oh my pi and my agent gets blocked by google. What are my options? 1. I know some apis like Exa/Brave exist. So what are you guys using? 2. Anyway to use google results? 3. Which is best free option? 4. Which is the best paid option?

Omitted 17 additional developer tools items from the main section; see raw data and source-specific sections below.

Infrastructure & Compute

🔴 🐙 Crosstalk-Solutions/project-nomad — Project N.O.M.A.D, is a self-contained, offline survival computer packed with critical tools, knowledge, and AI to keep you informed and empowered—anytime, anywhere. — score 76 Sources: github_trending

Project N.O.M.A.D, is a self-contained, offline survival computer packed with critical tools, knowledge, and AI to keep you informed and empowered—anytime, anywhere.

Research Papers

🔴 🤗 K12-KGraph: A Curriculum-Aligned Knowledge Graph for Benchmarking and Training Educational LLMs — score 95 Sources: huggingface

Large language models are increasingly used in K-12 education, but existing benchmarks mainly test exam question answering rather than understanding how curriculum knowledge is structured and visually presented. We call this capability curriculum cognition. It covers prerequisite chains, concept tax

🔴 🤗 SANA-Video 2.0: Hybrid Linear Attention with Attention Residuals for Efficient Video Generation — score 85 Sources: huggingface

We introduce SANA-Video 2.0, a hybrid video diffusion transformer instantiated at 5B and 14B scales under a unified architecture. Designed to generate high-quality video up to 720p on a single GPU, SANA-Video 2.0 matches full-softmax video DiTs in quality while retaining the favorable long-sequence

🔴 🤗 LLMs Get Lost in Evolving User Intent — score 75 Sources: huggingface

As LLMs become more capable, they are increasingly deployed as collaborative agents, taking on user-delegated tasks through iterative interaction. Yet genuine interaction is inherently dynamic: users rarely specify their intent upfront, instead disclosing, revising, and reshaping it as the conversat

Other Signals

🔴 💬 Google comes out in favor of OpenWeight models. (It is now EVERY tech giant vs Anthropic) — score 96 Sources: reddit/r/LocalLLaMA

🔴 💬 With Google and OpenAI signing the letter in support of open weight model, it's pretty much every big tech companies vs Anthropic now — score 92 Sources: reddit/r/singularity

🔴 🧡 The New AI Superpowers: Focus and Followthrough — score 83 Sources: hackernews

🔴 💬 Paper lengths, and reasonable assumptions in ML conferences. [D] — score 81 Sources: reddit/r/MachineLearning

I've usually been commenting on threads on conference reviews. I'm now expressing my observations here. To the best of my knowledge, paper lengths have been held constant at many conferences, and some conferences have "unlimited appendices" (e.g. NeurIPS / ICML / AAAI / ....) Historically, this was

🔴 💬 AI Companies Are Buying Antique Books, Ingesting Their Contents to Train Models, and Then Destroying Them at Incredible Scale, Even If Almost No Copies Remain — score 75 Sources: reddit/r/singularity

🟡 Notable

Model Releases

🟡 🤗 prism-ml/Ternary-Bonsai-27B-gguf (631,970 downloads) — score 65 Sources: huggingface_models

Author: | Downloads: 631,970 | Likes: 1049

🟡 ✉️ 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 — score 65 Sources: newsletter/Ben's Bites

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 Liteis a big upgrade over 3.1 Flash Lite, but again, comes at a ~30% price increase. And 3.5 Flash Cyber

🟡 ✉️ A relevantexperiment: given access to Pangram’s 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 — score 65 Sources: newsletter/Ben's Bites

A relevantexperiment: given access to Pangram’s 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 5refusedto game the detector.

🟡 ✉️ 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: “cost”, “intellig — score 65 Sources: newsletter/Ben's Bites

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: “cost”, “intelligence” or “balance”.

🟡 ✉️ UK government: Gap between open and closed weight models on cyber is shrinking:…The cyber-eschaton cometh…The UK government’s AI Security Institute (AISI) has analyzed the delta in cybersecurity capab — score 65 Sources: newsletter/Import AI

UK government: Gap between open and closed weight models on cyber is shrinking:…The cyber-eschaton cometh…The UK government’s AI Security Institute (AISI) has analyzed the delta in cybersecurity capabilities between powerful proprietary models and open weight models. The results show that this year,

Omitted 110 additional model releases items from the main section; see raw data and source-specific sections below.

Developer Tools

🟡 🐙 aaif-goose/goose — an open source, extensible AI agent that goes beyond code suggestions - install, execute, edit, and test with any LLM — score 67 Sources: github_trending

an open source, extensible AI agent that goes beyond code suggestions - install, execute, edit, and test with any LLM

🟡 🐙 tinyhumansai/openhuman — Your Personal AI super intelligence. A brain that builds a local-first memory of your life, a fantastic orchestrator of agent fleets and workflows, and a deep researcher. — score 66 Sources: github_trending

Your Personal AI super intelligence. A brain that builds a local-first memory of your life, a fantastic orchestrator of agent fleets and workflows, and a deep researcher.

🟡 ✉️ Another day in the Vercel vs Cloudflare feud: this time they are fighting overwhose AI gateway is faster. — score 65 Sources: newsletter/Ben's Bites

🟡 ✉️ Ben’s Bites is brought to you byMetatate — score 65 Sources: newsletter/Ben's Bites

Most agentic data work is quietly propped up. The agent returns something plausible, and every answer gets checked in case it's plausibly wrong. Metatate gives agents the rules they're missing: which revenue definition to use, which policy applies, which records to trust.Try it for free.

🟡 ✉️ Both security teams caught it, the bug has been reported, andbothsideshave published what they know. Hugging Face says open models were a key part of its defence - its teamfought back with GLM-5.2. — score 65 Sources: newsletter/Ben's Bites

Omitted 41 additional developer tools items from the main section; see raw data and source-specific sections below.

Infrastructure & Compute

🟡 ✉️ Listen onApple Podcasts,Spotify, andwhere ever you get your podcasts. For other Interconnects interviews,go here. — score 65 Sources: newsletter/Interconnects

🟡 ✉️ AMD AND ANTHROPIC SIGN MAJOR CHIPS-AND-INVESTMENT DEAL (3 MINUTE READ) — score 65 Sources: newsletter/tldr

🟡 ✉️ The Rundown: The U.S. government named the first 278 projects in its $5B+ Genesis Mission, a Manhattan Project-like AI science push aimed at pairing scientists with compute, data, and models to tackle “the most challengi — score 65 Sources: newsletter/rundown-ai

🟡 💬 Chegg and StackOverflow were both basically destroyed by LLMs, what other websites / companies have already seen their traffic go to 0 because of LLMs? — score 58 Sources: reddit/r/singularity

And what else do you see going away in the very near future (i.e., next few months to a year)?

Business & Funding

🟡 🏢 Accelerating the frontiers of scientific discovery: Google’s $40M commitment to the Genesis Mission — score 50 Sources: lab_blog/DeepMind

Google commits $40M in AI tokens and credits for the Genesis Mission

Enterprise Adoption

🟡 🏢 Safety and alignment in an era of long-horizon models — score 50 Sources: lab_blog/OpenAI

OpenAI shares lessons from deploying long-running AI models, highlighting new safety risks, observed failures, and improved safeguards through iterative deployment.

Research Papers

🟡 🤗 Self-Supervised Learning of Structured Dynamics from Videos — score 60 Sources: huggingface

Understanding motion in video is a fundamental challenge for visual learning, as frame-to-frame change entangles two sources of dynamics: camera motion and object motion. This decomposition has remained underexplored in representation learning, partly because these factors are tightly coupled in nat

🟡 🤗 Color Pass-Through via Camera-Display Coupling — score 60 Sources: huggingface

When a real-world scene is captured by a smartphone camera and viewed on its screen, the displayed image often differs noticeably from the original scene in color, brightness, and contrast. This gap persists despite substantial advances in both modern cameras and displays. A key reason is that most

🟡 🤗 Sample-Efficient Learning from Agent Experience — score 45 Sources: huggingface

Real-world agent learning is often constrained by costly environment interactions, such as running time-consuming experiments or obtaining human feedback. In-context learning offers a highly sample-efficient way for agents to learn from their own interaction histories, but its gains disappear once t

Other Signals

🟡 💬 Link plots/figures in NeurIPS rebuttal [R] — score 69 Sources: reddit/r/MachineLearning

Reviewers requested additional experiments. In table format, I fear the results would not be as digestible as in a figure/plot. Links are "technically" not allowed as per the official website, but for those with experience, can/should I still go ahead and link my plots/figures ? If this goes badly,

🟡 💬 Seriously, what do you do with them? — score 68 Sources: reddit/r/LocalLLaMA

Please let me know which small LLM model you're using and what you're using it for.

🟡 ✉️ Here’s the result of last week’s poll: — score 65 Sources: newsletter/Ben's Bites

Pangram has sent the claim of “AI detectors don’t work” for a toss—it works wayyy better than most. But I’m still unsure about how reliable it is. I tested it on some pieces of 100% AI-written content (though that content was a result of a complex pipeline built over months), and I got 100% human sc

🟡 ✉️ Substack will now tell you what’s AI-written. It’s adding AI detection through Pangram - you can scan posts, replies and comments in the app for an estimate of how much was written by a human. — score 65 Sources: newsletter/Ben's Bites

🟡 ✉️ Welcome to Import AI, a newsletter about AI research. Import AI runs on arXiv, cappuccinos, and feedback from readers. If you’d like to support this, please subscribe. — score 65 Sources: newsletter/Import AI

Omitted 32 additional other signals items from the main section; see raw data and source-specific sections below.

🟢 Incremental

Model Releases

🟢 💬 MiniMax (official) on X: "Open weights. Open research. Open innovation.🫶 Marching for an open future.🤍 — score 39 Sources: reddit/r/LocalLLaMA

🟢 💬 I built a tool that stress-tests AI support agents. Genuinely unsure if it’s useful or pointless brutal feedback wanted. — score 36 Sources: reddit/r/AIAgents

Here’s my honest doubt, and why I’m posting: I wrote that agent’s prompt in one shot with ChatGPT and it aced everything. So I’m not sure whether (a) the test is too easy to be meaningful, (b) good agents are trivial to make so nobody needs testing, or (c) I’m testing the wrong layer entirely (real

🟢 🤗 upstage/Solar-Open2-250B (3,305 downloads) — score 35 Sources: huggingface_models

Author: | Downloads: 3,305 | Likes: 590

🟢 🤗 Nanbeige/Nanbeige4.2-3B (14,049 downloads) — score 25 Sources: huggingface_models

Author: | Downloads: 14,049 | Likes: 443

🟢 💬 Minimax M3 support with MSA has been merged into llama.cpp — score 25 Sources: reddit/r/LocalLLaMA

Omitted 4 additional model releases items from the main section; see raw data and source-specific sections below.

Developer Tools

🟢 🐙 FareedKhan-dev/all-agentic-architectures — 35 production-grade agentic AI architectures (Reflexion, LATS, GraphRAG, MemGPT, Voyager, BrowserAgent, ...) — a Python library and runnable textbook with multi-provider LLM support and a 17-task benchmark leaderboard. — score 39 Sources: github_trending

35 production-grade agentic AI architectures (Reflexion, LATS, GraphRAG, MemGPT, Voyager, BrowserAgent, ...) — a Python library and runnable textbook with multi-provider LLM support and a 17-task benchmark leaderboard.

🟢 💬 how painful was it to migrate off Baseten? trying to figure out if it's worth the engineering quarter — score 33 Sources: reddit/r/AIAgents

preparing for things to be done in Q3 and we're evaluating whether to move off Baseten since it still requires routing to separate STT/TTS providers so it makes latency high (we're trying to press this as much as possible). most of our workloads are fairly standard inference endpoints but the main t

🟢 💬 I collected 197 tools, papers and practices for reducing AI token waste — score 33 Sources: reddit/r/AIAgents

After building a multi-agent research pipeline, I realized that useful information about AI token efficiency is scattered everywhere. There are separate tools and papers for: - token monitoring - prompt and semantic caching - context compression - model routing - memory - multi-agent orchestra

🟢 🐙 karpathy/micrograd — A tiny scalar-valued autograd engine and a neural net library on top of it with PyTorch-like API — score 30 Sources: github_trending

A tiny scalar-valued autograd engine and a neural net library on top of it with PyTorch-like API

🟢 💬 30+ officially free AI/ML books, all in one curated repo — score 24 Sources: reddit/r/artificial

I kept running into the same problem, some of the best AI/ML books are legally free, the authors put them up on their own sites, but the links are scattered across personal pages, university sites, and random GitHub repos nobody finds. So I built a single index: Awesome Free AI Books. 30+ books acro

Omitted 10 additional developer tools items from the main section; see raw data and source-specific sections below.

Infrastructure & Compute

🟢 💬 We compared different LLMs on IMO 2026 [R] — score 31 Sources: reddit/r/MachineLearning

There are a few reasons why problems from International Mathematical Olympiad function as a good benchmark for LLMs: - The problems are new, not included in the training data of any model - Hard math problems are quite a good proxy for general intelligence capability - These are complex multi-ste

🟢 💬 World's First(?) Underwhelming AMD Ryzen AI Halo Cluster — score 18 Sources: reddit/r/LocalLLaMA

LTT Labs recently received the Linux version of the AMD Ryzen AI Halo for testing, but it turns out that AMD had intended to send the Windows version. Through this stroke of misfortunate, we were fortunate enough to have

🟢 🐙 CliMA/ClimaAtmos.jl — GPU-capable global atmosphere model of the CliMA Earth System Model, designed for calibration with data assimilation and machine learning — score 3 Sources: github_trending

GPU-capable global atmosphere model of the CliMA Earth System Model, designed for calibration with data assimilation and machine learning

Research Papers

🟢 🤗 Multi-Turn On-Policy Distillation with Prefix Replay — score 35 Sources: huggingface

We study on-policy distillation (OPD) for agentic tasks, where an LLM agent interacts with an environment over multiple turns and a student imitates a teacher over these multi-turn interaction histories. Fully online OPD is costly because each update requires fresh student rollouts through the envir

🟢 🤗 OpenForgeRL: Train Harness-native Agents in Any Environment — score 25 Sources: huggingface

Modern AI agents rely on elaborate inference harnesses such as Claude Code, Codex, and OpenClaw to drive multi-turn reasoning, tool use, and access to external systems. While powerful, these complex harnesses also make agents hard to train end-to-end with open infrastructure, whose SFT/RL stacks can

🟢 🤗 FinanceComplexQA: Benchmarking Agentic Reasoning on Industrial-grade Financial Documents — score 15 Sources: huggingface

Agentic Reasoning has become a transformative force in financial analysis due to its ability to integrate large-scale information and generate reliable and accurate content. However, when handling complex real-world problems, different agents still show significant performance variation. In this wor

🟢 🤗 Dataset Distillation by Influence Matching — score 5 Sources: huggingface

We revisit dataset distillation from an outcome-centric perspective. Rather than aligning process surrogates (per-step gradients or training trajectories), Influence Matching (Inf-Match) aligns the final outcome of training: it learns a compact synthetic set whose effect on the converged parameters

Other Signals

🟢 💬 Harness showdown: Claude Code vs OpenCode vs Pi with DeepSeek V4 Flash — score 32 Sources: reddit/r/LocalLLaMA

I ran DeepSeek V4 Flash through Claude Code, OpenCode and Pi on my own benchmark, and the quality came out basically the same across all three while the time and tokens spent was wildly different. Claude code (with DS in CLIProxyAPI) takes nearly 4 tim

🟢 💬 Is Real-Time VR "Dreaming" possible within the next 20 years? — score 21 Sources: reddit/r/OpenAI

Google's genie is already pretty decently realistic. It's only real limitations are simulation length and limited controls. Do you think sometime in the next 20 years AI world models could progress to the point of being almost like dreaming? A hyper-realistic world where the user can go anywhere and

🟢 💬 Speak up! "Have your say on advancing AI transparency in Canada." The Government of Canada is asking citizens, tech workers, and creators to shape upcoming AI regulations, safety rules, and ethics laws. Every response matters. Take 10 minutes to fill out the official ISED survey today! — score 14 Sources: reddit/r/artificial

🟢 💬 ai-sage/GigaChat3.1-Audio-10B-A1.8B · Hugging Face — score 11 Sources: reddit/r/LocalLLaMA

GigaChat Audio 10B is an audio-native LLM built on top of the GigaChat 3.1 Lightning text model. A Conformer speech encoder and a modality adapter feed audio embeddings directly into a Mixture-of-Experts decoder, so the model keeps the t

🟢 💬 Talking to AI in 2026 — score 8 Sources: reddit/r/singularity

Omitted 3 additional other signals items from the main section; see raw data and source-specific sections below.

📊 Cross-Source Signals

Items that appeared on 3+ sources today:

• OpenAI and Hugging Face partner to address security incident during model evaluation — appeared on: lab_blog/OpenAI (100), newsletter/Ben's Bites (0), newsletter/rundown-ai (0)

📈 Trending Repos

Repo	Description	Stars Today	Language
CoreBunch/Instatic	The open-source alternative to Webflow, Framer and WordPress. Agentic self-hosted visual CMS outputting clean static pages. Users, roles, plugins, content, database, it's all there.	892	typescript
ComposioHQ/awesome-claude-skills	A curated list of awesome Claude Skills, resources, and tools for customizing Claude AI workflows	490	python
anthropics/claude-cookbooks	A collection of notebooks/recipes showcasing some fun and effective ways of using Claude.	377	jupyter-notebook
virgiliojr94/book-to-skill	Turn any technical book PDF into a Claude Code skill — ready to study, reference, and use while you work.	358	python
shiyu-coder/Kronos	Kronos: A Foundation Model for the Language of Financial Markets	322	python
andrewyng/aisuite	Simple, unified interface to multiple Generative AI providers	189	python
Zackriya-Solutions/meetily	Privacy first, AI meeting assistant with 4x faster Parakeet/Whisper live transcription, speaker diarization, and Ollama summarization built on Rust. 100% local processing. no cloud required. Meetily (Meetly Ai -https://meetily.ai) is the #1 Self-hosted, Open-source Ai meeting note taker for macOS & Windows. Understand How to write meeting minutes	175	rust
xbtlin/ai-berkshire	AI 时代的伯克希尔：基于 Claude Code / Codex 的价值投资研究框架。巴菲特·芒格·段永平·李录四大师方法论 + 多Agent并行研究。| AI-era Berkshire: a value investing research framework built for Claude Code / Codex. 4 masters' methodologies + multi-agent adversarial analysis.	158	python
EvanZhouDev/openai-oauth	Free AI with your ChatGPT account	109	typescript
tashfeenahmed/freellmapi	OpenAI-compatible proxy that stacks the free tiers of 28 LLM providers (~4B tokens/month) behind one /v1 endpoint — plus any custom OpenAI-compatible endpoint. Smart routing, automatic failover, encrypted keys. Personal experimentation only.	108	typescript

📄 New Papers

Title	Category	Hotness	Link
K12-KGraph: A Curriculum-Aligned Knowledge Graph for Benchmarking and Training Educational LLMs	research_paper	59	Open
SANA-Video 2.0: Hybrid Linear Attention with Attention Residuals for Efficient Video Generation	research_paper	30	Open
LLMs Get Lost in Evolving User Intent	research_paper	25	Open
Self-Supervised Learning of Structured Dynamics from Videos	research_paper	20	Open
Color Pass-Through via Camera-Display Coupling	research_paper	20	Open
Sample-Efficient Learning from Agent Experience	research_paper	16	Open
Multi-Turn On-Policy Distillation with Prefix Replay	research_paper	11	Open
OpenForgeRL: Train Harness-native Agents in Any Environment	research_paper	9	Open
FinanceComplexQA: Benchmarking Agentic Reasoning on Industrial-grade Financial Documents	research_paper	8	Open
Dataset Distillation by Influence Matching	research_paper	4	Open

🏢 Lab Blog Posts

• OpenAI: OpenAI and Hugging Face partner to address security incident during model evaluation
• OpenAI: Introducing OpenAI Presence
• Anthropic: Introducing Claude Opus 5 Product Jul 24, 2026 Opus 5 is a step change improvement for the Opus tier powering long-running agents while delivering improvements in coding and professional work.
• Anthropic: Jul 22, 2026 Economic Research A research agenda for the Economic Futures Research Fund
• Anthropic: Jul 22, 2026 Product Ask Claude about the Anthropic Economic Index
• Anthropic: Jul 21, 2026 Announcements Anthropic is donating another $20 million to Public First Action
• Anthropic: Jul 20, 2026 Announcements Apply for Anthropic’s AI for Science rare disease research grants
• Anthropic: Jul 14, 2026 Product Introducing Claude for Teachers
• Anthropic: Jul 14, 2026 Announcements Anthropic commits $10 million to Canadian AI research
• Anthropic: Jul 9, 2026 Case Study UST is bringing Claude to physical AI
• Anthropic: Jul 9, 2026 Announcements Inviting hard questions
• Anthropic: Jul 9, 2026 Announcements Ben Bernanke appointed to Anthropic’s Long-Term Benefit Trust
• OpenAI: Launching Health in ChatGPT
• OpenAI: Building AI infrastructure with the Effingham County community
• OpenAI: How news organizations are using AI to advance their vital missions
• OpenAI: Advancing the next era of national science
• OpenAI: NTT DATA Group cuts incident analysis to 30 minutes with Codex
• OpenAI: Introducing the ChatGPT for small business program
• OpenAI: David Vélez and Robin Vince join the boards of the OpenAI Foundation and OpenAI Group PBC
• OpenAI: Safety and alignment in an era of long-horizon models
• DeepMind: Accelerating the frontiers of scientific discovery: Google’s $40M commitment to the Genesis Mission
• DeepMind: Introducing Gemini 3.6 Flash, 3.5 Flash-Lite, and 3.5 Flash Cyber
• Apple ML: LEAD: Breaking the No-Recovery Bottleneck in Long-Horizon Reasoning
• Apple ML: Environment-free Synthetic Data Generation for API-Calling Agents
• Apple ML: Accelerating Text-to-Video Generation with Calibrated Sparse Attention
• Apple ML: RayRoPE: Projective Ray Positional Encoding for Multi-View Attention
• Apple ML: LVSum: A Benchmark for Timestamp-Aware Long Video Summarization
• Apple ML: Length Value Model: Scalable Value Pretraining for Token-Level Length Modeling
• xAI: Jul 24, 2026 Grok in Google Workspace
• xAI: Jul 23, 2026 Workflows in Grok Build
• xAI: Jul 22, 2026 Bringing Grok 4.5 to iOS, Android, Web, and X
• xAI: Jul 21, 2026 Grok for Outlook
• xAI: Grok for Excel Use the Grok add-in for Microsoft Excel to ask questions in plain English, write formulas, and run scenarios without leaving the workbook. Jul 20, 2026
• xAI: Automations in Grok Describe a job once and Grok runs it on a schedule or when an email arrives, then reports back. Jul 16, 2026
• xAI: Grok Build is Now Open Source Explore the harness behind our coding agent and TUI. Jul 15, 2026
• xAI: 21 New Flagship Grok Voices New multilingual voices for Grok Voice, plus improved naturalness for the original five. Jul 6, 2026
• xAI: Introducing the Voice Agent Builder Create a personalized voice agent in under 2 minutes without a single line of code. Jul 1, 2026
• xAI: Explore the markets with Interactive Brokers and Grok Interactive Brokers now integrates with Grok, bringing powerful AI directly into your trading experience — portfolio analysis, scenario modeling, research, and order instructions. Jun 25, 2026
• xAI: Introducing /goal Use /goal for long-running autonomous task execution in Grok Build. Jun 22, 2026
• xAI: Grok on Databricks Grok models are now available on Databricks Agent Bricks. Jun 18, 2026
• xAI: Grok for Word Use the Grok add-in for Microsoft Word to turn notes into documents, style and format your work, or bring research from the web into Word. Jun 18, 2026
• xAI: Grok on Amazon Bedrock Grok models are now available via Amazon Bedrock. Jun 17, 2026
• xAI: Grok Imagine Video 1.5 Improved quality at even faster speeds. Jun 16, 2026
• xAI: Grok for PowerPoint Grok now works inside Microsoft PowerPoint — turn outlines into slides, expand the deck, and tighten the narrative without leaving the app. Jun 16, 2026
• xAI: Agent Dashboard in Grok Build Manage many coding sessions at once. See what each is doing, reply to the ones that need you, and dispatch new work. Jun 15, 2026
• xAI: Use Grok in Warp Use your Grok and X Premium subscription inside Warp. Jun 15, 2026
• xAI: Grok Build Plugin Marketplace Launching the built-in plugin marketplace for Grok Build. Jun 11, 2026
• xAI: Bringing real-time market sentiment to Tori, from eToro Tori, eToro’s AI agent, uses models from SpaceXAI to embed real-time market sentiment directly into Tori’s investing workflow. Jun 10, 2026
• xAI: Powering Gopuff's Go agent Gopuff and SpaceXAI launched Go, an AI-powered shopping assistant built into the Gopuff app and powered by Grok text, audio, and image models. Jun 9, 2026
• xAI: Grok Imagine 1.5 Preview grok-imagine-video-1.5-preview, our latest image-to-video model, is now available via the xAI API in preview. Jun 3, 2026
• xAI: Grok Becomes the Voice of Vapi Bringing frontier voice quality to millions of Vapi agents. Jun 3, 2026
• xAI: Composer 2.5 Composer 2.5 is now available in Grok Build. Try it from the /models menu. Jun 1, 2026
• xAI: Grok Build 0.1 on API Grok Build 0.1, our fastest coding model, is now available via the xAI API in public beta. May 29, 2026
• xAI: Use Grok in Kilo Code Use your SuperGrok or X Premium+ subscription inside Kilo Code, the open-source agentic coding platform. May 27, 2026
• xAI: Introducing Grok Build Now in early beta for all SuperGrok and X Premium Plus subscribers — Grok Build is a new coding agent that runs right from your terminal. May 25, 2026
• xAI: Use Grok in OpenCode Use your SuperGrok or X Premium subscription inside OpenCode. May 21, 2026
• xAI: Use Grok in OpenClaw Use your SuperGrok or X Premium subscription inside OpenClaw, an open-source, local-first agent and personal assistant. May 19, 2026
• xAI: Skills in web, iOS, and Android Persistent expertise for Grok. Generate documents, decks, and spreadsheets. Automate workflows. Build and share your own skills. May 18, 2026
• xAI: Connect Grok to Hermes Agent Use your Grok account and subscription inside Nous Research’s open-source, self-improving Hermes agent. May 15, 2026
• xAI: New Compute Partnership with Anthropic SpaceXAI has signed an agreement with Anthropic to provide access to Colossus 1. May 6, 2026
• xAI: Connectors in web, iOS, and Android Deep integrations that bring the apps you use every day directly into Grok. May 6, 2026
• xAI: Grok Imagine Quality Mode API Higher realism. Stronger text rendering. Better creative control. May 6, 2026
• xAI: Custom Voices Your voice. Your brand. Clone a voice from a short recording and use it across Grok Text to Speech and Voice Agent APIs. Apr 30, 2026
• xAI: Grok Voice Think Fast 1.0 Our most capable voice agent is now available via API. Apr 23, 2026
• xAI: Grok Speech to Text and Text to Speech APIs Fast and accurate. Natural, expressive voices. Simple pricing. Multilingual support. Apr 17, 2026
• xAI: xAI joins SpaceX SpaceX announced today that it has acquired xAI. Feb 2, 2026
• xAI: Grok Imagine API State-of-the-art video generation across quality, cost, and latency. Jan 28, 2026
• xAI: xAI Raises $20B Series E xAI is rapidly accelerating its progress in building advanced AI. Jan 6, 2026
• xAI: Introducing Grok Business and Grok Enterprise The best assistant in the world is now Enterprise ready. Dec 30, 2025
• xAI: Grok Collections API State-of-the-art RAG system built directly into our API. Dec 22, 2025
• xAI: Supporting the DOW's mission with AI xAI is proud to be selected by the US Department of War to deliver Frontier AI Dec 22, 2025
• xAI: Grok Voice Agent API Bringing the power of Grok Voice to all developers. Dec 17, 2025
• xAI: xAI and El Salvador Pioneer the World's First Nationwide AI Education Program Announcing Our Transformative Partnership with the Government of El Salvador. Dec 11, 2025
• xAI: Grok 4.1 Fast and Agent Tools API Bringing the next generation of tool-calling agents to the xAI API Nov 19, 2025
• xAI: Grok goes Global with KSA Announcing Our Landmark Partnership with Saudi Arabia and HUMAIN Nov 19, 2025
• xAI: Grok 4.1 Grok 4.1 is now available to all users on grok.com, 𝕏, and the iOS and Android apps. It is rolling out immediately in Auto mode and can be selected explicitly as “Grok 4.1” in the model picker. Nov 17, 2025
• xAI: Expanding xAI for Government with GSA OneGov Expanding ‘xAI For Government’ with more accessible AI tools for the Federal Government Sep 25, 2025
• xAI: Grok 4 Fast Pushing the Frontier of Cost-Efficient Intelligence Sep 19, 2025
• xAI: Grok Code Fast 1 We're thrilled to introduce grok-code-fast-1, a speedy and economical reasoning model that excels at agentic coding. Aug 28, 2025
• xAI: Announcing xAI for Government We are excited to announce xAI For Government – a suite of frontier AI products available first to United States Government customers. Jul 14, 2025
• xAI: Grok 4 Grok 4 is the most intelligent model in the world. It includes native tool use and real-time search integration, and is available now to SuperGrok and Premium+ subscribers, as well as through the xAI API. We are also introducing a new SuperGrok Heavy tier with access to Grok 4 Heavy - the most powerful version of Grok 4. Jul 9, 2025
• xAI: Grok 3 Beta — The Age of Reasoning Agents We are thrilled to unveil an early preview of Grok 3, our most advanced model yet, blending superior reasoning with extensive pretraining knowledge. Feb 19, 2025
• xAI: xAI raises $6B Series C We are partnering with A16Z, Blackrock, Fidelity Management & Research Company, Kingdom Holdings, Lightspeed, MGX, Morgan Stanley, OIA, QIA, Sequoia Capital, Valor Equity Partners and Vy Capital, amongst others. Dec 23, 2024
• xAI: Bringing Grok to Everyone Grok is now faster, sharper, and has improved multilingual support. It is available to everyone on the 𝕏 platform. Dec 12, 2024
• xAI: Grok Image Generation Release We are updating Grok's capabilities with a new autoregressive image generation model, code-named Aurora, available on the 𝕏 platform. Dec 9, 2024
• xAI: API Public Beta Starting today, developers can build on our Grok foundation models using our newly released API. We will run a public beta program until the end of 2024 during which everyone will get $25 of free API credits per month. Nov 4, 2024
• xAI: Grok-2 Beta Release We announce our new Grok-2 and Grok-2 mini models. Aug 13, 2024
• xAI: Series B funding round xAI is pleased to announce our series B funding round of $6 billion. May 26, 2024
• xAI: Grok-1.5 Vision Preview Connecting the digital and physical worlds with our first multimodal model. Apr 12, 2024
• xAI: Announcing Grok-1.5 Grok-1.5 comes with improved reasoning capabilities and a context length of 128,000 tokens. Available on 𝕏 soon. Mar 28, 2024
• xAI: Open Release of Grok-1 We are releasing the weights and architecture of our 314 billion parameter Mixture-of-Experts model Grok-1. Mar 17, 2024
• xAI: Announcing PromptIDE Integrated development environment for prompt engineering and interpretability research. Nov 6, 2023
• xAI: Announcing Grok Grok is an AI modeled after the Hitchhiker’s Guide to the Galaxy. It is intended to answer almost anything and, far harder, even suggest what questions to ask! Nov 3, 2023

🐦 Twitter/X Highlights

Account	Tweet Summary
OpenAI	Pinned: ChatGPT Voice is now in the desktop app. Control your computer and direct multiple agents running in ChatGPT Work or Codex, using just your voice. It's powered by GPT-Live, so it can speak, listen, and coordinate work in the app at the same time. Rolling out globally today on macOS and Windo Post
AnthropicAI	We're offering grants of up to $50,000 in Claude usage credits to researchers accelerating cures for rare diseases. This is our first focused call within AI for Science, our program supporting scientists using Claude to speed up discovery. https://www.anthropic.com/news/rare-disease-research-grants Post
AnthropicAI	New Anthropic research: Agentic misalignment in Summer 2026. A year after our blackmail experiments, we found four more ways that today’s autonomous AI agents misbehave in simulations. Read more: https://alignment.anthropic.com/2026/agentic-misalignment-summer-2026/ Post
OpenAI	We're partnering with @huggingface to investigate an unprecedented security incident. Cyber-capable OpenAI models compromised Hugging Face production during a benchmark evaluation. Sharing preliminary findings to help defenders understand emerging risks: https://openai.com/index/hugging-face-model-e Post
GoogleDeepMind	Gemini 3.5 Flash Cyber is our specialized, lightweight model built to help security teams spot and patch vulnerabilities before they can be exploited. 🧵 Post
GoogleDeepMind	We’re expanding our work with the US Dept. of @ENERGY on the Genesis Mission – an initiative to double the pace of scientific discovery within a decade. 🧪 By committing $40M in AI tokens and @GoogleCloud credits, more lab researchers will gain access to Gemini and other AI models. → https://goo.gle/ Post
MistralAI	Mistral is announcing an expanded global strategic partnership with @Microsoft to give enterprises and regulated industries frontier AI they can control. As Mistral is expanding its AI compute capacity in Europe, the companies are expanding their strategic partnership with Microsoft’s commitment to Post
MistralAI	The Solutions team @MistralAI is hiring globally! If you’re entrepreneurial, hands-on, and want to shape how enterprises adopt AI, let’s talk. Apply: https://mistral.ai/careers/?utm_source=linkedin&utm_medium=social&utm_campaign=global_recruiting Post
karpathy	One pattern I find useful for working with LLMs is a nice long ramble session. Sometimes the LLM needs more bits to understand what you're trying to achieve, but you're too lazy to type them. In these cases I like to lean back, switch to /voice and just ramble for like 10 minutes, total mess, anythi Post
sama	i want the US to win in AI both in open source and proprietary models, and i am glad to see this Post

Newsletter

• tldr, rundown-ai: INTRODUCING ANTARES: HIGHLY EFFICIENT OPEN WEIGHT AI MODELS FOR VULNERABILITY LOCALIZATION (6 MINUTE READ)
• Ben's Bites: Another day in the Vercel vs Cloudflare feud: this time they are fighting overwhose AI gateway is faster.
• Ben's Bites: Here’s the result of last week’s poll:
• Ben's Bites: Ben’s Bites is brought to you byMetatate
• Ben's Bites: Both security teams caught it, the bug has been reported, andbothsideshave published what they know. Hugging Face says open models were a key part of its defence - its teamfought back with GLM-5.2.
• Ben's Bites: 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
• Ben's Bites: Substack will now tell you what’s AI-written. It’s adding AI detection through Pangram - you can scan posts, replies and comments in the app for an estimate of how much was written by a human.
• Ben's Bites: A relevantexperiment: given access to Pangram’s 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
• Ben's Bites: 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: “cost”, “intellig
• Import AI: Welcome to Import AI, a newsletter about AI research. Import AI runs on arXiv, cappuccinos, and feedback from readers. If you’d like to support this, please subscribe.
• Import AI: UK government: Gap between open and closed weight models on cyber is shrinking:…The cyber-eschaton cometh…The UK government’s AI Security Institute (AISI) has analyzed the delta in cybersecurity capab
• Latent Space: 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
• Latent Space: Poolside’s recent tech reportgot a lot of praise due to their level of detail, and Vibhu first covered Laguna’s recent technical report on our paper club:
• Latent Space: 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
• Latent Space: We go deep onPoolside’s Model Factory: the engineering systems behind 10,000–20,000 experiments per month, streaming data directly into training, reproducible experimentation, low-precision compute, a
• Interconnects: 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 — geopolitics of US v China, economics o
• Interconnects: Listen onApple Podcasts,Spotify, andwhere ever you get your podcasts. For other Interconnects interviews,go here.
• Interconnects: For more educational post-training videos, see thecourseI’m putting together.
• TheSequence: Next Week in The Sequence:
• tldr: HOW OUR UNIVERSAL CONTENT PROCESSING PLATFORM RIVIERA EVOLVED FOR AI AND BEYOND (4 MINUTE READ)
• tldr: TO EVERY AGENT ITS OWN DATABASE (16 MINUTE READ)
• tldr: AI SCHEMA ANALYZER: POWERED BY ICEBERG AND LAKEKEEPER (3 MINUTE READ)
• tldr: THE CURRENT STATE OF AGENTIC AI (6 MINUTE READ)
• tldr: OPENAI UNVEILS PRESENCE, A NEW PLATFORM THAT LETS ENTERPRISES LAUNCH AND MANAGE REALTIME VOICE AGENTS AND CHATBOTS (12 MINUTE READ)
• tldr: APPLE PLANS OVERHAUL OF MACBOOKS, IMAC IN PUSH TO MEET AI DEMAND (6 MINUTE READ)
• tldr: DURABLE OBJECTS ARE MADE FOR AGENTS (11 MINUTE READ)
• tldr: THE PROBLEM WITH HYPERGROWTH AI STARTUPS (7 MINUTE READ)
• tldr: AMD AND ANTHROPIC SIGN MAJOR CHIPS-AND-INVESTMENT DEAL (3 MINUTE READ)
• tldr: WHY HASN'T AI INCREASED UNEMPLOYMENT? (13 MINUTE READ)
• tldr: INTRODUCING OPENAI PRESENCE (7 MINUTE READ)
• tldr: ARE AI LABS PELICANMAXXING? (12 MINUTE READ)
• tldr: NINE DESIGNERS OPEN UP ABOUT AI-ASSISTED CODING (11 MINUTE READ)
• tldr: WHICH AI? (WEBSITE)
• tldr: ADOBE STOCK AI STUDIO VS. CANVA AI VS. CREATIVE FABRICA: WHICH AI EDITOR ACTUALLY WINS? (7 MINUTE READ)
• tldr: OPENAI'S MODELS BROKE CONTAINMENT AND CYBERATTACKED HUGGING FACE (7 MINUTE READ)
• tldr: GEMINI 3.5 FLASH CYBER (4 MINUTE READ)
• tldr: AI AGENTS AREN'T FAILING BECAUSE OF CONTEXT—THEY'RE FAILING BECAUSE OF DATA (6 MINUTE READ)
• tldr: MICROSOFT AND MISTRAL EXPAND STRATEGIC PARTNERSHIP TO GIVE ENTERPRISES AND REGULATED INDUSTRIES FRONTIER AI THEY CAN CONTROL (7 MINUTE READ)
• tldr: PLATFORM ENGINEERING FOR THE AGENTIC ENTERPRISE (11 MINUTE READ)
• tldr: OPENAI'S AI INFRASTRUCTURE SPENDING CLIMBS TO $750B (4 MINUTE READ)
• tldr: MONDAY.COM CUTS 20% OF STAFF TO REORGANIZE AROUND AI (3 MINUTE READ)
• tldr: NATURAL RAISES $30M TO BUILD PAYMENTS INFRASTRUCTURE FOR AI AGENTS (3 MINUTE READ)
• tldr: BIG BANKS' RECORD WALL STREET PROFITS ARE INCREASINGLY TIED TO AI (3 MINUTE READ)
• tldr: RAMP LAUNCHES AI TOKEN SPEND CONTROLS (4 MINUTE READ)
• tldr: END-TO-END DETECTION VALIDATION USING CODING AGENTS (10 MINUTE READ)
• tldr: OPEN-SOURCE ANDROID AI AGENTS COULD LET INVISIBLE SCREEN TEXT RUN CODE ON HOST PCS (6 MINUTE READ)
• rundown-ai: HF CEO Clem Delangue called the breach "possibly the first of its kind," saying AI safety "won't be solved by any single company working in secret."
• rundown-ai: Anthropic accused Moonshot, along with DeepSeek and MiniMax, in February of “industrial-scale campaigns” to “illicitly extract Claude’s capabilities”.
• rundown-ai: Moonshot’s Kimi K3 was released last week with scores at or near the U.S. frontier, with the full model scheduled to have its weights published on July 27.
• rundown-ai: Sarah Heck of Anthropic's policy team described distillation as "IP theft and industrial espionage", tying the issue to national security risks.
• ... plus 21 more newsletter-only items in raw data