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 20 new AI releases, including Claude Sheets and Whiteboards — Anthropic adds office apps inside Claude, Claude designs electronics — three working devices from short prompts and Sam Witteveen — 'Gemini 4 Carbon is Live!'. 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
- Claude Sheets and Whiteboards — Anthropic adds office apps inside Claude
Anthropic launched Claude Sheets and Claude Whiteboards in beta on all paid plans, made Claude Data (formerly Dashboards) generally available, and let partner apps like Notion and Atlassian open as full apps inside Claude.
- Claude designs electronics — three working devices from short prompts
Anthropic and Diode Computers report that Claude Fable 5.2 designed a speakerphone, a voice-assistant hub and a handheld virtual pet end to end, and all three boards were made straight from its files with no manufacturing back-and-forth.
- Sam Witteveen — 'Gemini 4 Carbon is Live!'
Sam Witteveen walks through Google's Gemini 4 Carbon on the day it opened to developers: what it is built for, its benchmarks, pricing and cost per task against Claude Fable 5.2, and three app-building demos.
- Gemini 4 Carbon — now open to developers, Ultra users and Search
Gemini 4 Carbon is now available to paying developers in Google AI Studio and Gemini for business, and to AI Ultra users in the Gemini app and AI Mode. It first shipped on 28 October to cyber defenders only.
- Gemini Robotics 2.5 — DeepMind's most capable robot action model
Gemini Robotics 2.5 is Google DeepMind's most capable vision-language-action model. It was tested on 200+ dexterous tasks, with success rates 30% to 100% higher than Gemini Robotics 2. It reached 96% on a Mercedes-Benz kitting pilot.
- Gemini for healthcare — Google Cloud's agent for clinical and payer work
Gemini for healthcare is a Google Cloud agent solution that links health records, claims and admin systems to do multi-step work like prior authorization. It is in U.S. preview, with HCA Healthcare, Highmark Health and Northwell Health.
- AlphaProof Code — DeepMind AI verifies the Signal protocol's code
AlphaProof Code is a DeepMind agent framework that writes Lean proofs that code is correct. It verified the Signal protocol's implementation in four weeks, and DeepMind is open-sourcing the verification artifacts.
- Two Minute Papers — 'Is Software Over?'
Two Minute Papers asks 'Is Software Over?' and links four projects: MirrorCode, where AI agents rebuild programs from behaviour alone, the REMEND decompiler, WebUI-95 and PhotoCraft, an open-source Photoshop rebuild in Rust.
- LongCat-2.5 — Meituan opens weights of its 1.6T agentic model
Meituan released the weights of LongCat-2.5 under the MIT license. It is a 1.6T-parameter multimodal mixture-of-experts model with about 48B active per token, built for long-horizon agent tasks.
- Gemini human review — excerpts from your Gmail and Drive can reach reviewers
A public write-up says consumer Gemini with Connected Apps can pull Gmail and Drive content into chats that human reviewers may see. Google's own help page confirms reviewers see excerpts from relevant emails and files.
- e-means — Meta FAIR's k-means replacement needs one-tenth the epochs
e-means is a mini-batch replacement for k-means clustering from Meta FAIR. It reaches the loss Lloyd's k-means gets in 25 epochs in 1.4 to 2.6 epochs, and runs 10-45x faster on GPU. The code ships in Faiss.
- Claude Code 2.1.322 — fast mode comes to Google Vertex AI
Claude Code 2.1.322 adds fast mode on Google Vertex AI in the global, us and eu regions. When Vertex refuses a fast request, Claude Code falls back to standard speed instead of failing.
- Wes Roth — 'NOTHING will be smarter than Fable 5.2...'
Wes Roth reviews Claude Fable 5.2 and Mythos 5.2, focusing on Anthropic's note that it limited Fable 5.2's training on kernel optimization and LLM architecture research, and tests Fable 5.2 on his own game benchmark.
- TPU 8tx and TPU 8i Max — Google adds a confidential TPU and a dual-chip TPU
Google adds two eighth-generation TPU systems. TPU 8tx runs confidential inference on TPU 8t silicon with AMD EPYC hosts. TPU 8i Max puts two TPU 8i chips in one package for twice the compute, SRAM and HBM bandwidth.
- Fireworks AI breach widens — customer datasets and model weights accessed
Fireworks AI says attackers reached its development machines through a hacked cloud provider on October 24. They accessed billing data for customers and, for a limited set of accounts, secrets, datasets and incomplete model weights.
- OpenAI Control Room — a fail-closed off switch for research model workloads
OpenAI Control Room tracks every research inference job and can stop one within 35 seconds. Inference engines ask it before each request, and if the service goes down, all research inference is denied.
- Fireship — 'OpenAI spent $1.2 million hacking Linux... and it was a bargain'
Fireship breaks down OpenAI's Linux kernel audit, in which an unreleased OpenAI model came back with 948 root exploits for about $1.2 million in API credits.
- Muse Ember — Meta's smallest Muse model for agent sub-tasks
Muse Ember is Meta's smallest and most efficient Muse model, built for tool calls, coding subtasks, routing and extraction. It costs $0.10 per million input tokens and $0.50 per million output tokens, with a 1M-token context.
- Claude Code 2.1.321 — Fable 5.2 as the default Fable model, MCP OAuth fix
Claude Code 2.1.321 adds Claude Fable 5.2 as the default Fable model, lets remote-control apply one settings file to every session, and fixes MCP OAuth sending refresh tokens to another host after a redirect.
- Claude Fable 5.2 — Anthropic's new top model, $8/$40 and 30% faster
Claude Fable 5.2 is Anthropic's new top-end model for coding, agents and knowledge work. It scores 69.9% on FrontierSWE v2 against 56.3% for Fable 5.1, costs $8 per million input tokens and $40 per million output tokens, and is up to 30% faster.
- llama-modchip 2.4 — 284B DeepSeek-V4-Flash on a 64 GB CPU, no GPU
llama-modchip patches llama.cpp to run the 284B DeepSeek-V4-Flash MoE model in under 59 GiB of RAM at 256K context with no GPU. Version 2.4.0 raises CPU prefill by 77% to 44.36 tok/s.
- Arete-27B — Nous Research strips state propaganda from Qwen3.8-27B
Arete-27B is Nous Research's open-weight fine-tune of Qwen3.8-27B that cuts state-aligned answers on political topics from 22% to 0.9% on its 803-prompt test, with no loss on Terminal-Bench-2.1.
- Sam Witteveen — 'NVIDIA's Secret 2 Bit Model' on Nemotron 3.5 Bolt 74B-A9B
Sam Witteveen explains how NVIDIA's Nemotron 3.5 Bolt 74B-A9B squeezes into about 24GB with 2-bit experts, covering ParetoQ, pruning and quantization-aware training, then tests it locally on agentic tool calls.
- Claude Code 'Don't Ask Again' — passwords leak in 1,122 public GitHub files
Apiiro found passwords in 1,122 public GitHub files saved by Claude Code's 'Don't Ask Again' option. Approved commands land in .claude/settings.local.json with any secret typed inline, and some repos committed that file.
- OpenAI hires Hacktron — the team that hacked it now tests its systems
OpenAI is partnering with Hacktron, the AI security startup that reached its internal code through the bug bounty, to find vulnerabilities in OpenAI's systems, starting with connections to third-party services.
- Microsoft Research Closes AI Frontiers — the group behind Autogen
Microsoft Research's AI Frontiers group is ending, John Langford writes. The roughly 50-person team behind Autogen and the Fara agent was folded into Microsoft AI in September, and its leaders have left.
- Two Minute Papers — 'OpenAI's Astra Just Opened The Black Box Of AI'
Two Minute Papers covers Weight-Space Programming, the MIT CSAIL study where GPT-6 Astra reads raw neural network weights and edits them by hand, with no training loop.
- Sam Witteveen — 'Hy4 Beats it's Preview Easily'
Sam Witteveen looks at the full release of Tencent's Hunyuan Hy4, which follows the August preview, and says its results beat the preview. The open weights are still pending.
- BioBasicsBench — OpenAI tests AI agents on everyday wet-lab data
BioBasicsBench is a 100-task OpenAI benchmark, built with Ligo, that asks AI agents to read raw lab data, catch errors and flag weak evidence. GPT-6 Astra leads at 66.7%; GPT-5.2 scores 31.2%.
- Meta Active Memory and Active Focus — Muse Wonder trains the Muse app models
Meta is rolling out specialized versions of the models behind Muse and Meta AI, trained using Muse Wonder. They add Active Memory and Active Focus, and Muse now scores 47% on Meta's Personal Superintelligence Index, up from 38%.
- Ajeya Cotra — 'AI is dramatically accelerating AI research right now'
Ajeya Cotra argues that METR's report on Anthropic shows AI R&D has already crossed the 'dramatic acceleration' threshold, and that if the speedup compounds, a naive extrapolation points to superintelligence within a year.
- Components of agent memory — Anthropic tests how Claude remembers
Anthropic's engineering post splits agent memory into four layers and tests them. Memory lifted Claude Opus 5.5 from 0.655 to 0.838 on CL-Bench, and offline 'dreaming' fixed wrong notes from past transcripts.
- How researchers use Claude — Anthropic maps AI across 207,049 science sessions
Anthropic's Economics team studied 207,049 science sessions on Claude and Claude Code and surveyed 528 scientists. Data analysis led use, and the team estimates current AI could take on 27–52% of researcher time.
- Hightouch — a Jev that asks Opus for help matches Opus at 4x lower cost
Hightouch tested Jev and GPT-6 Luna as cheap judges of identity resolution. A Jev that sends unsure cases to Claude Opus 5.5 scored 98.8% against Opus's 99.0%, at $0.88 versus $3.86 per 1,000 identities.
- Claude Code 2.1.319 — MCP tasks run in the background, deny instead of prompt
Claude Code 2.1.319 moves MCP tool calls that run as tasks to the background, adds a setting that denies a tool call instead of waiting on a permission prompt, and tightens auto mode behind LLM gateways.
- Fireship — 'OpenAI's nameless model just changed the odds of P = NP...'
Fireship explains OpenAI's claim that an unnamed model wrote an algorithm that solves 3-SAT faster than exponential time, which would break the Exponential Time Hypothesis.
- Two Minute Papers — 'The Future of AI Is Absolutely Incredible (Spencer Huang)'
Two Minute Papers interviews NVIDIA product manager Spencer Huang about why robots are hard, training data, simulated worlds and the gap between simulation and reality, in a 46-minute conversation.
- Wes Roth — 'Anthropic just built it...' on METR's AI R&D audit
Wes Roth walks through METR's independent assessment of AI R&D acceleration at Anthropic, why much of the audit was redacted, and what the AECI benchmark trend says about the pace of Claude's progress.
- Sam Witteveen — 'AREX-2 - The Harness in the Model'
Sam Witteveen explains AREX-2, BAAI's 27B open-weight model fine-tuned from Qwen3.8-27B that runs the try, check and retry loop itself instead of leaving it to the agent harness.
- MCP Explained — an interactive guide to every MCP message on the wire
MCP Explained is a free, animated guide from Runlayer that steps through the Model Context Protocol one JSON-RPC message at a time, from lifecycle and transports to auth, tools and known attacks. It hit 279 points on HN.