Claude Code 2.1.239 fixes a Bedrock proxy bug that double-billed you
It's the second recent Claude Code release to quietly fix a cost bug — and OpenAI just cut GPT-5.6 Sol's price and added per-request regions.
Copy markdown
Bedrock behind a proxy? You were paying twice
Claude Code 2.1.239 fixes streaming through proxies that strip the response Content-Type header — the client was silently re-running every turn non-streaming, doubling your billed API calls. Update, then eyeball recent Bedrock spend.
Prompt caching was silently off through your gateway
The related fix from 2.1.237: prompt caching now works for sessions on an LLM gateway or custom base URL. If you route Claude Code through one, you'd been paying full, uncached token prices on every call.
One command off anthropic 0.x
`/claude-api upgrade` migrates Python projects from the anthropic 0.x SDK to 1.x. The v1.0 that landed Aug 20 moved to httpx2 and needs Python 3.10+; timeouts now use `anthropic.Timeout`, not `httpx.Timeout`.
GPT-5.6 Sol just got cheaper — until November
OpenAI cut GPT-5.6 Sol to $4/M input and $20/M output — 20% off input, 33% off output — through Nov 21. It applies to the API and Codex-plan credits; standard subscriptions are unchanged.
Route an OpenAI request to a specific region
Projects with Global geography can now send an individual request to a chosen region via a prefixed domain — per-request data residency without standing up a separate project.