AI/TLDR — every new AI model, tool, repo & paper
The latest AI releases, refreshed every 2 hours and explained in plain English.
What AI shipped today?
In the last 24 hours AI/TLDR tracked 13 new AI releases, including Checkpoint forensics — Chinese open models carry a Blackwell GPU fingerprint, smolmaxxing — a 16B MoE model runs on a 1 GB TV box and a 2013 iPad and Claude Code 2.1.323 — Code Mode lets Claude call its tools from JavaScript. AI/TLDR is an AI release tracker that follows new AI models, open-source tools, papers, datasets and benchmarks — refreshed every 2 hours from verified primary sources and explained in plain English.
AI Release Index — live stats on AI releases · Learn AI
- Checkpoint forensics — Chinese open models carry a Blackwell GPU fingerprint
Will DePue's checkpoint forensics reads the GPU layout left in open model weights. He finds a Blackwell pattern in six models from DeepSeek, Moonshot, MiniMax and Ant Group, including DeepSeek-V4-Flash and Kimi K3.
- smolmaxxing — a 16B MoE model runs on a 1 GB TV box and a 2013 iPad
smolmaxxing is an MIT llama.cpp patch for 1 GB devices. It runs Qwen3.5-2B at 7.7 tokens/s and the 16B Ling-mini-2.0 MoE at 5.1 tokens/s on a Xiaomi TV box, and 1-bit Bonsai-1.7B at 24–26 tokens/s on a 2013 iPad Air.
- Claude Code 2.1.323 — Code Mode lets Claude call its tools from JavaScript
Claude Code 2.1.323 adds Code Mode, an experimental opt-in where Claude writes JavaScript to call its tools. Unknown model IDs now get a 1M-token context window by default instead of 200K.
- Fireship — 'Nvidia just got caught with its pants down...'
Fireship breaks down Jensen Huang's resignation as Nvidia CEO and the scandal behind it, including a Singapore company that bought $4.8 billion of Nvidia GPUs.
- Google Intelligent Eyewear — Gemini glasses open for preorder from $395
Google Intelligent Eyewear, built with Samsung, Warby Parker and Gentle Monster, is open for preorder. The Gemini glasses start at $395, go on sale December 10 and work with Pixel, Galaxy and iPhone phones.
- VS Code 1.146 — the Local agent harness is deprecated, Copilot harness stays
VS Code 1.146 deprecates the Local agent harness, which will be removed in February 2027, so agent sessions move to the Copilot harness. It also adds an Inbox for agent sessions, skills from MCP servers and a /code handoff from Copilot CLI.
- Gemini 4 Hydrogen — Google's fast, low-cost model for agent work
Gemini 4 Hydrogen is Google's fast model for coding and agent workflows. It costs $1.50 in and $7.50 out per 1M tokens, has a 1M-token context window, and scores 51 on the Artificial Analysis Intelligence Index, one point above Gemini 4 Argon.
- Codex Security 0.10.0 — OpenAI's scanner can resume scans and cap costs
Codex Security 0.10.0 lets interrupted scans resume from a checkpoint, adds a shared cost budget for bulk scans, imports ArmorCode SAST findings, and explains why a blocked patch failed. The repo is trending on GitHub today.
- Two Minute Papers — 'DeepMind's AI Just Did The Impossible'
Two Minute Papers covers AlphaProof Code, the DeepMind agent system that wrote over six million lines of Lean to formally verify the code behind the Signal protocol in four weeks.
- A2UI 1.0 — Google's generative UI protocol reaches its first stable version
A2UI 1.0 is the first stable version of Google's open protocol for generative UI. An agent describes a screen, and each app draws it with its own native components. 1.0 adds two-way function calls, accessibility and new agent SDKs.
- ha-mcp 9.2.0 — the Home Assistant MCP server trends on GitHub
ha-mcp is an MIT-licensed MCP server with 87 tools that lets Claude, ChatGPT and other assistants build and debug a Home Assistant setup. Version 9.2.0 moves to Python 3.14 and hands helper flows to Home Assistant Core.
- ERNIE 5.5 Flash Preview — Baidu's 280B MoE for agentic coding
ERNIE 5.5 Flash Preview is Baidu's first post-trained model on ERNIE 5.5 Flash Base: 280B total and 14B active parameters, 1M-token context. It is on Qianfan and free on OpenRouter; Apache-2.0 weights are promised.
- MACaQ — Huawei open-sources a 4-bit quantizer for GGUF, vLLM and Diffusers
MACaQ is Huawei CSL's Apache-2.0 quantizer that calibrates 4-bit weights and activations one layer at a time. Its Qwen3.6-35B-A3B GGUF is closer to FP16 than Unsloth's file of the same size.
- Yogurt — a free, open-source Granola alternative that runs on your computer
Yogurt is a free, MIT-licensed meeting-notes app for Mac, Windows and Linux. It records calls without a bot, then transcribes and summarizes them with AI models that run on your own computer.
- agent-browser 0.42 — Vercel's agent browser CLI can record and replay flows
agent-browser 0.42.0, Vercel Labs' Rust browser CLI for AI agents, adds experimental saved flows that record a browser task once and replay its requests. The 49.9k-star repo is on GitHub's daily trending list.
- Weight-space programming as a linear program — 8x fewer edits than GPT-6 Astra
Nathan Barry shows that GPT-6 Astra's weight edit to Qwen3.5-0.8B can be found as a linear program. Two one-minute LP solves match Astra's scores while changing 16,291 weights instead of 130,188.
- Voltage glitching unlocks an A100 — a mining card gets 80 GB and full compute
A researcher used a voltage fault on NVIDIA's GA100 chip to blow locked fuses and turn a CMP 170HX mining card into an A100-80G, unlocking about 80 GB of HBM and up to 126 SMs. NVIDIA was told before publication.
- MiniMax H3-Image — open-weight model makes 4K images in under 10 seconds
MiniMax H3-Image is an open-weight image model built on H3. It generates and edits 16 MP (4096×4096) images in 6 sampling steps, takes up to 16 reference images, and is #2 on Hugging Face trending.
- Jensen Huang resigns as Nvidia CEO — CFO Colette Kress is interim CEO
Nvidia says Jensen Huang has resigned as president, CEO and board member, effective immediately. CFO Colette Kress is interim CEO while the board searches for a permanent one, and Harvey C. Jones becomes Nvidia's first board chair.
- United Airlines — bans third-party AI agents from booking flights
United Airlines says it will block unauthorized third-party AI agents from booking flights for travelers. It cites account and payment security, unintended purchases, and possible federal rule violations.
- AWS Emulator — a free local copy of the AWS APIs, built by AI agents
AWS Emulator is a free, CC0-licensed Rust program that answers AWS CLI and SDK calls on your own machine, with no AWS account. AWS says AI agents wrote all of its code and its engineers did not review it.
- Gemini Decisions API — fast, calibrated answers to closed-set questions
Google's Gemini Decisions API runs the new gemini-decision-1 model to score yes/no, multiple-choice and rubric questions in one forward pass, with confidence scores, 150-250 ms latency and $0.10 per 1M input tokens.
- ByteDance plans a Muse-like AI assistant — launch in weeks, Bloomberg says
ByteDance plans to launch an AI assistant like Meta's Muse within weeks, Bloomberg reports. Co-founder Zhang Yiming reportedly told staff ByteDance cannot match OpenAI and Anthropic at the frontier before 2029 or 2030.
- Tabletop — Google Labs shared spaces for group plans and games
Tabletop is a Google Labs experiment where a group shares one space and prompts @tabletop to build tools like itineraries, polls, expense splitters and brackets. It is a U.S. waitlist for ages 18+.
- 48 AI agents scan public lidar — and flag an unregistered Maya site
Pietro Miotti, who says he is not an archaeologist, had 48 AI agents review lidar images of about 1,200 km² of Maya-region jungle. They flagged a hilltop site with a plaza and an oval pyramid that Mexico's INAH says is not registered.
- AlphaGenome 2 — DeepMind's cell-type-aware DNA effect model
AlphaGenome 2 is a Google DeepMind model that predicts how DNA variants change gene activity in a chosen human cell type. It trained on 100M+ single-cell profiles across 68 tissues and is free for non-commercial research.
- Qwen4-Max-Preview — Alibaba opens a two-week preview of its next flagship
Qwen4-Max-Preview is the first Qwen4 model anyone can use: a multimodal flagship with a 1M-token context, served only through Alibaba's Token Plan from November 11 to 25. No benchmarks, weights or per-token price yet.
- Claris Harbor — Apple's Claris tests a private AI server for Macs
Claris Harbor, from Apple's Claris unit, turns Apple silicon Macs into a private AI server for open-weight models. Several Macs can be linked over Thunderbolt 5 to pool memory. It is in private preview.
- Amazon swaps Claude Sonnet 4.6 for Strands Decider — ticket labels 98% cheaper
Two Amazon engineers replaced Claude Sonnet 4.6 with the Strands Decider E4B decision model for labeling internal tickets. Calls got 78% faster, and cost fell from $0.0222 to about $0.0005 per ticket.
- Lex Toumbourou — Claude Fable 5.2 one-shots a 12-stage beat 'em up
Lex Toumbourou used Claude Code with Claude Fable 5.2 to build his nephew's game design, Aiden's Revenge, in one overnight run: 12 stages, 18 characters and four-player controller support, using about 340M Claude tokens.
- VulnCheck on Glasswing — 442 CVEs, 2 exploited, maintainers carry the triage
VulnCheck's Patrick Garrity counts 442 CVEs credited to Anthropic's Project Glasswing after seven months. Only 2 are exploited in the wild, and he argues the OSS Scanner shifts triage work onto open-source maintainers.
- ChatGPT Enterprise on AWS Marketplace — buy it on your AWS bill
OpenAI now sells ChatGPT Enterprise through AWS Marketplace. Companies buy it with a private offer, pay on their AWS bill, and eligible spend counts toward AWS EDP or PPA commitments. Codex is part of the package.
- AI Explained — 'Fable 5.2: The Self-Improvement Loop Keeps Loopin’'
AI Explained's 10 November 2026 video reads the 264-page Claude Fable 5.2 report and METR's 30-page report on AI speeding up AI research, and tests Fable 5.2 and Gemini 4 Carbon.
- Businesses in ChatGPT — claim your listing and run a Business Agent
OpenAI launched Businesses in ChatGPT in beta. Companies claim their profile in Business Hub, see mentions and clicks, and US ecommerce and online-service firms can build a Business Agent that answers customers inside ChatGPT.
- Anthropic IPO terms — $70 to $78 a share, $26B from seven chip and cloud firms
Anthropic launched its IPO roadshow: 825 million shares at $70 to $78 on Nasdaq as ANT. A new filing adds a $26B share sale to AMD, Broadcom, Google, Micron, Microsoft, Nvidia and TSMC at the IPO price.
- Evan X. Merz — AI usage generates a fragile codebase
Evan X. Merz, who wrote at least 90% of his code with AI at Meta, argues that AI-built SaaS codebases get fragile: shallow tests, over-complex code, weaker human understanding and more pull requests than reviewers can check.
- Matteo — I'm not convinced the AI bubble exists
Matteo argues the AI bubble is not a settled fact: most of the roughly $5 trillion in AI spending is one-time data-center building, and Anthropic's S-1 shows $2.08 of revenue per $1 of compute. He still lists real risks.
- Modern Web Guidance 1.0 — Chrome skills that modernize agent-written web code
Modern Web Guidance 1.0 is an Apache-2.0 skill pack and offline CLI from the Chrome team. It gives coding agents current web guidance for 250 features, so they stop writing legacy JavaScript and CSS.
- Gemini Intelligence November update — Android widgets from a prompt
Gemini Intelligence on Android adds Create My Widget, which builds a home-screen widget from a plain description, plus Device Assistance for fixing phone settings by asking. Pixel 10 phones get the features in the November Pixel Drop.
- Fireship — 'Did Anthropic just save its $1 trillion IPO?'
Fireship looks at Claude Fable 5.2 and Claude Mythos 5.2, which Anthropic released in the middle of its IPO roadshow, and at what the new models can do.