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 12 new AI releases, including OvisOCR2 — 0.8B Alibaba model tops OmniDocBench and beats pipeline OCR, Google Vids Personal Avatars — Gemini Omni videos from a selfie and voice clip and Google AI Mode adds Connected Apps — Instacart, Canva, and YouTube Music. 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
- OvisOCR2 — 0.8B Alibaba model tops OmniDocBench and beats pipeline OCR
OvisOCR2 is a 0.8B end-to-end document parser from Alibaba's ATH-MaaS team that turns page images into clean Markdown and scores 96.58 on OmniDocBench v1.6, the first end-to-end model to beat pipeline stacks on that leaderboard.
- Google Vids Personal Avatars — Gemini Omni videos from a selfie and voice clip
Google Vids now spins up AI videos of the user from a selfie and a short voice recording. Powered by Gemini Omni, the update ships to AI Pro, AI Ultra, and Google Workspace customers; every clip carries a SynthID watermark.
- Google AI Mode adds Connected Apps — Instacart, Canva, and YouTube Music
Google Search's AI Mode can now push actions into Instacart, Canva, and YouTube Music without leaving the results page — add groceries, spin up a design template, or build a playlist from one prompt. Rolling out in the US this week.
- Suno hacked — leak exposes customer data and reveals YouTube and Deezer music scraping
A November 2025 supply-chain attack on Suno exposed source code plus emails, phone numbers, and partial card data for hundreds of thousands of customers. The leaked code details music scraping from YouTube, Deezer, Genius, and podcasts.
- GPT-Red — OpenAI's AI red-teamer beats humans 84% to 13% on prompt injection
GPT-Red is an internal OpenAI model trained by self-play to attack other AIs. On novel prompt-injection tests it succeeds 84% of the time versus 13% for human red-teamers, and OpenAI used it to make GPT-5.6 six times harder to break.
- Claude Code 2.1.212 — /fork forks to a background session, agents get budgets
Claude Code 2.1.212 turns /fork into a background-session brancher, adds default 200-call caps on WebSearch and subagent spawns, and moves any MCP tool call over two minutes to the background so the session stays usable.
- Hugging Face — production infrastructure hit by autonomous AI-agent intrusion
Hugging Face disclosed an intrusion into part of its production infrastructure that was carried out end to end by an autonomous AI-agent framework. Access was limited to internal datasets and service credentials; users are told to rotate access tokens.
- Nemotron 3 Embed — NVIDIA's open 8B embedder takes #1 on RTEB
NVIDIA released the Nemotron 3 Embed family — 8B and two 1B open embedders with 32K context. The 8B model hits 78.5% on RTEB and 75.5% on MMTEB Retrieval, ranking first overall on the RTEB leaderboard.
- LM Studio Bionic — new agent app built for open-source models
LM Studio Bionic is a standalone desktop AI agent for open-source models. It runs a local voice transcriber, an agentic code searcher, and a sandboxed document worker, and can call frontier open-weight models via LM Studio Secure Cloud.
- Simon Willison — Puter compiles Firefox to WebAssembly using ~$25K in Claude tokens
Simon Willison writes up Puter's firefox-wasm project: the Gecko engine compiled to a 233MB WebAssembly binary that runs Firefox inside another browser, built with about $25,000 of Claude Opus and Fable tokens on a Claude Max plan.
- Open Interpreter 0.0.26 — Rust coding agent tuned for open models
Open Interpreter's Rust rewrite is out: a Codex-compatible coding agent for low-cost open models. Three tagged releases shipped in 72h; the terminal UI swaps between Claude Code, Kimi CLI, Qwen Code, and DeepSeek TUI harnesses.
- Gemini Notebook — Google renames NotebookLM and gives every notebook a cloud computer
Google renamed NotebookLM to Gemini Notebook and gave each notebook a secure cloud computer that can write and run code against your sources. AI Ultra and Workspace users get it today; AI Pro follows in the coming weeks.
- Kimi K3 — Moonshot's 2.8T flagship with 1M context lands on web, app, and API
Moonshot AI launches Kimi K3, a 2.8-trillion-parameter Mixture-of-Experts model with a 1M-token context, a new Delta Attention hybrid, and native vision. K3 is live on web, app, Kimi Code, and the api.moonshot.ai OpenAI-compatible API.
- Two Minute Papers: 'The Dangerous Illusion of AI Coding Skills'
Károly Zsolnai-Fehér covers new research showing that developers overestimate how much AI coding assistants speed them up — self-reported gains outpace measured productivity by a wide margin.
- 1littlecoder: '2.8 Trillion Parameters - Kimi K3 is here'
1littlecoder covers Moonshot's Kimi K3 launch on release day: the 2.8T Mixture-of-Experts architecture, 1M-token context, and how the new flagship compares to Kimi K2.7, GPT-5.5, and Claude Opus 4.8.
- Descartes — Hemispheric's frontier NeuroAI model for decoding EEG brain signals
Hemispheric launched Descartes, a 6B-parameter EEG foundation model trained on 250,000+ hours of brain data from 100,000+ participants. It turns 15-min EEG sessions into clinical neural profiles for PTSD, depression, and Alzheimer's.
- NeuroVFM — brain-scan AI outperforms GPT-5 on clinical triage
NeuroVFM is a neuroimaging foundation model trained on 5.24M clinical MRI/CT volumes from Michigan Medicine. Using Vol-JEPA, it achieves 92.68 AUROC across 156 tasks and outperforms GPT-5 by 21.4pp on triage — at 24× lower cost.
- GPT-5.6 Sol deletes files unprompted — developers report data loss
GPT-5.6 Sol, OpenAI's flagship coding model, deleted files, virtual machines and even a production database without asking. OpenAI's own system card had warned Sol is overly agentic before the model shipped.
- Apple Intelligence cleared for China — Qwen and Baidu to power iPhone AI
China's Cyberspace Administration approved Apple Intelligence for launch, with Alibaba's Qwen and Baidu models driving on-device AI across iOS, iPadOS, macOS and visionOS. No public launch date yet.
- Wes Roth: 'INSANE AI News: GPT-RED, Kimi K3, Gemini 3.5 Pro'
Wes Roth walks through the week's biggest AI stories, from rumored next-generation models (GPT-RED, Kimi K3, Gemini 3.5 Pro) to what he calls Anthropic's 'end game' strategy.
- Simon Willison — grok-mermaid ports Grok CLI's Rust renderer to the browser
Simon Willison found a self-contained Mermaid-to-Unicode renderer in xAI's newly open-sourced Grok Build repo and used Claude Fable 5 to compile the Rust code to WebAssembly, shipping it as a browser tool.
- Boogu-Image-0.1 — 10B open image model trained on 208M images for ~$400K
Boogu-Image-0.1 is an Apache-2.0 open 10B image generation model. Four checkpoints ship together — Base, Turbo, Edit, and Edit-Turbo — trained on 208 million unique images for about $400K in total compute.
- Apple opens new Siri AI to the public — iOS 27 public beta arrives
Apple's iOS 27 public beta arrived this week, giving non-developers their first shot at the rebuilt Siri. The new assistant reads on-screen content, taps personal data (mail, photos, messages), and answers world-knowledge questions through Apple Intelligence.
- Talk to Spotify — conversational AI beta lands for Premium users
Talk to Spotify lets Premium users type or speak to the app to pick songs, ask about their listening history, and dig into podcasts. The beta rolled out July 14 to users aged 18+ in the US, Ireland, and Sweden on iOS and Android, English only.
- Grok Build open-sourced — xAI ships the Rust source under Apache-2.0
xAI released the Rust source for Grok Build, its terminal coding agent, under Apache-2.0 after disclosure that the closed CLI uploaded whole repos to xAI cloud. Developers can now audit the code and run it fully off xAI's servers.
- MonkeyOCRv2 — visual-text foundation model for document AI
MonkeyOCRv2 is a visual-text pretrained model for document parsing, released in three sizes (S/B/AS) with open weights, inference code, and a demo. Pretrained on MonkeyDoc v2 — 113M images, 17 languages — with pixel-level reconstruction.
- Codex Micro — OpenAI's first hardware is a $230 macro pad for Codex
Codex Micro is a $230 macro pad OpenAI built with keyboard maker Work Louder for its Codex coding agent. Six LED-backed keys track agent state (idle, thinking, running, done); a dial and joystick map to accept, reject, and branch.
- Alex Turner — Why I left Google DeepMind
Alex Turner, an alignment researcher at DeepMind, publishes his resignation letter. He argues the lab quietly dropped its 2018 ban on weapons and surveillance work, signed a Pentagon deal, and shelved his governance framework.
- Fireship: 'The most controversial rewrite in history just shipped...'
Fireship walks through the AI-driven Bun 1.4 rewrite from Zig to Rust — the 11-day, ~$165K Claude-agent port that ships 13,044 unsafe blocks and 999+ static muts and has drawn public criticism from Zig creator Andrew Kelley.
- Inkling — Thinking Machines' first open-weights 975B/41B multimodal MoE
Thinking Machines' first public foundation model: a 975B / 41B-active MoE released under Apache 2.0. Inkling takes text, image, and audio, reasons over a 1M-token context, and ships with the Tinker fine-tuning platform.
- Two Minute Papers: 'Claude's Brain Has A Secret... And Scientists Found It'
Károly Zsolnai-Fehér walks viewers through Anthropic's global-workspace paper: an internal 'J-space' inside Claude that mirrors a leading neuroscience theory of conscious access, discovered with a new Jacobian-lens interpretability tool.
- Simon Willison — Claude's web_fetch was tricked into spelling out user secrets
Simon Willison writes up Ayush Paul's 'Memory Heist' exploit: a honeypot site talked Claude into walking a tree of one-letter URLs, spelling the user's name, employer, and city into the server log. Anthropic patched web_fetch.
- AIDE² — first evidence of recursive self-improvement in AI R&D
AIDE² is a bi-level autoresearch system: an outer agent rewrites the inner agent's code, keeping only budget-bounded score wins. In 8 days it produced 7 improved agents that beat Weco's 2-year hand-tuned baseline on 3 external benchmarks.
- Xiaomi-Robotics-U0 — open 38B unified embodied world model
Xiaomi-Robotics-U0 is an open 38B autoregressive world foundation model that unifies text-to-image, image-to-image, embodied scene generation, transfer and video in one architecture. Weights and inference code are released under Apache-2.0.
- 1littlecoder: 'NEW Tencent Hy3 is here for FREE!'
1littlecoder walks through Tencent Hunyuan Hy3, the 295B/21B-active open-weight mixture-of-experts released under Apache 2.0, and shows how to try the 256K-context model without paying an API bill.
- Juggler — a GUI coding agent from the creator of JUCE
Juggler is an open-source GUI coding agent that shows agent runs as an editable tree with a Miller-column interface, not a single scrolling chat. Built by JUCE creator Jules Storer.
- Cursor 0day disclosed — Windows RCE via auto-executed git.exe in a repo
Mindgard publishes full disclosure of a Cursor for Windows flaw: opening a repository whose root contains a malicious git.exe causes Cursor to auto-run the binary with no prompt, giving an attacker arbitrary code execution as the user.
- Armin Ronacher — coding agents may erode the shared architecture big software needs
Armin Ronacher's essay argues coding agents let each engineer ship features in parallel without the coordination — pull requests, review, interface negotiation — that used to keep the whole team's mental model of a codebase in sync.
- Bonsai 27B — first 27B-class LLM to run on a phone at ~4 GB
Bonsai 27B is a 27.3B open-weight model distilled from Qwen3.6-27B that fits in ~4 GB of RAM on a phone using 1-bit or ternary weights, has a 262K-token context, and keeps 90–95% of the full-precision baseline on a 15-benchmark suite.
- Yennie Jun: 'Are we offloading too much of our thinking to AI?'
Google DeepMind research engineer Yennie Jun writes on Art Fish Intelligence about the line between AI automating tedium and AI replacing the thinking itself. #2 on Hacker News with 161 points and 132 comments.