Model Anthropic ships Claude Sonnet 5 Anthropic released Claude Sonnet 5, now the default model in Claude Code with a native 1M-token context window. Promotional pricing runs $2 and $10 per million tokens through August 31. For builders that means longer context and cheaper agent runs without moving up to a heavier model. anthropic |
Policy US lifts export controls on Fable 5 and Mythos 5 The US Department of Commerce lifted export controls on Claude Fable 5 and Mythos 5, and Anthropic is restoring access over the coming days. Teams that hit regional walls get the frontier models back online globally. A rare case of policy moving in the builder's favor. simonwillison |
Threat Intel Claude Code is steganographically marking its requests A researcher reports that Claude Code embeds a hidden steganographic signal in its outgoing requests. For threat intel and fraud professionals it is a live example of provenance marking baked directly into an AI coding tool. Worth reading for how invisible watermarking actually shows up in the wild. lobsters |
Workflow Claude Code v2.1.197: Sonnet 5 default plus 1M context Updating Claude Code to v2.1.197 makes Sonnet 5 the default and unlocks a native 1M-token context window inside the agent. Promotional pricing of $2 and $10 per million tokens holds through August 31. A one-command update that changes how much of a codebase you can hand an agent in a single pass. claude-code-releases |
Model Hindsight: open-source memory layer for agents (v0.8.4) Hindsight is an open-source memory layer for AI agents, now at v0.8.4. It gives agents persistent recall across sessions instead of starting cold on every run. Clone it to add durable memory to your own agent stack. gh-hindsight |
Workflow Hindsight ships an OpenHands integration (v0.1.1) Hindsight released a v0.1.1 integration for OpenHands, the open-source coding agent. It drops the persistent memory layer straight into an OpenHands workflow. A quick way to test agent memory inside a real coding-agent loop. gh-hindsight |
Model Nano Banana 2 Lite: Google's cheapest image model Google released Nano Banana 2 Lite, also known as Gemini 3.1 Flash Lite Image, its fastest and cheapest image model. It is live now in the Gemini API and built for high-volume generation. Worth a spin if you need cheap image output at scale inside an app. simonwillison |