Latest
Gemini Can Now Edit PDFs and Docs Without Leaving Chat
Google turned on in-chat file editing for Gemini on 29 October. What you can edit, the web-only limit on everything but PDFs, and the availability caveat.
GPT-4o Leaves the API January 14: Your Migration Plan
GPT-4o shuts off on the OpenAI API on 14 January 2027, with only three months' notice. Every date, the Azure auto-upgrade trap, and what to check first.
OpenAI Doubles Pro Codex Limits and Scraps the Pro 200 Cut
Work and Codex allowances double on Pro 100 and Pro 500 at the same prices, and the planned Pro 200 reduction is cancelled. Plus is unchanged.
Open Speech Models Beat Apple's Transcriber on Word Error Rate
A new Swift package swaps the engine behind Apple's speech-to-text API, and NVIDIA's Parakeet scored 1.43% word error rate against Apple's 2.12%.
Le Chonk's Weights Are Out: Should You Deploy Mistral ML4?
Mistral released the weights for its trillion-parameter Large 4 model, claiming parity with GLM-5.3 while using 47% fewer tokens per task.
Android Studio Rabbit 2 Opens to 40+ AI Coding Agents
Android Studio Rabbit 2 is stable with Bring Your Own Agent: 40+ coding agents in the IDE, Google's own speed numbers, and which APIs are still alpha.
MCP TypeScript SDK v2: What Changed and the One Gotcha
V2 splits the old single package and targets the July 2026 spec. The exact versions that work, the console.log trap, and when a custom server pays off.
Claude Code Limits Explained: Weekly, 5-Hour and Resets
Two ceilings run at once and share one pool with claude.ai. What drains them fastest, how to read /usage, and why the API overflow valve needs its own cap.
Genie 4 Is Gated and Unpriced: Who Should Actually Care
Genie 4 opened to trusted testers on October 27 with no price and no public API. What shipped, what drifts, and which teams can act on it now.
Apple Puts an AI Assistant and MCP Into Pages and Keynote
Apple is adding an AI assistant to Pages, Keynote and Numbers in beta, and the three apps now support the Model Context Protocol on macOS 27 or later.
GPT-6.1 Terra Costs $1/M Input: What Should Move to It
OpenAI's GPT-6.1 Terra ships at $1 per million input tokens with computer use built in. The full price table, the 272K cliff, and what to move.
Best LLM for Browser Agents: Sonnet 5.5 vs Opus 5.5 vs Holo4
Vendor-claimed OSWorld scores and list prices for six browser-agent models, plus the per-task math showing caching cuts the bill by over 80%.
One File in ~/Downloads Can Hijack Your Coding Agent
An AGENTS.md dropped into ~/Downloads by a web page made a coding agent install malware, reproduced against Codex with a success rate above 90%.
