OpenAI adds transparent PNG backgrounds to the gpt-image-2 API
No background-removal step for logos, icons or stickers — plus LLM 0.33 provider mixing, fresh Claude Code keybindings, and OpenRouter reasoning traces.
Copy markdown
Transparent backgrounds, straight from the API
gpt-image-2 now accepts background:"transparent" with PNG or WebP output in preview — logos, icons, stickers and UI assets come out ready to composite, with no background-removal pass. Build this weekend: a prompt-to-asset pipeline that drops clean, transparent PNGs straight into your design system.
LLM 0.33: mix providers, layer templates
Simon Willison's llm CLI hits 0.33 — moved to the OpenAI 3.x library and httpx2 (ending the install break 0.32.1 had to hot-patch), added per-call --key on llm embed/embed-multi, repeatable -t/--template to stack config and prompt, and a reasoning_summary (auto/concise/detailed) for Responses-API models. Upgrade to keep multi-provider scripts running.
Claude Code adds readline keybindings and header-auth marketplaces
Claude Code 2.1.238 ships keybindingFlavor for readline-style key bindings and a plugin-marketplace header-helper for dynamic HTTP headers, so authenticated or private marketplaces work; self-hosted runners get deferred shutdown, and 2.1.239–241 stack on reliability fixes. Worth an update if you run a self-hosted runner or a gated plugin marketplace.
Reasoning traces reach OpenRouter and Gemini in llm
The llm plugin ecosystem kept pace: llm-openrouter 0.7 adds reasoning traces and server-side tools, and llm-gemini 0.33 wires in Gemini 3.7 Flash. Route through OpenRouter or Gemini and you get the same trace-and-tool workflow core 0.33 unlocks — one CLI, one syntax, across providers.