After Anthropic introduced Claude Opus 5.5, Vercel added the model to AI Gateway, a service with a unified API for calling models. Vercel warns that two API changes can turn requests that were previously valid into HTTP 400 errors.
The model now chooses how much reasoning to use for each request, and requests that disable reasoning or set a fixed budget are rejected. Forced tool use is no longer available, so Vercel advises asking the model to use the tool and retrying missed calls in client code.
Claim check:
- Anthropic introduced Claude Opus 5.5 for full-cycle software development tasks. (confirmed by the publication itself: evidence; «The company introduced Claude Opus 5.5 on Tuesday to span more of the software application development lifecycle: from design specification creation through debugging to code generation and testing.»)
- Claude Opus 5.5 from Anthropic is available in Vercel AI Gateway. (confirmed by the publication itself: evidence; «Claude Opus 5.5 from Anthropic is now available on AI Gateway .»)
- Two API changes can turn previously valid requests into HTTP 400 errors. (confirmed by the publication itself: evidence; «Opus 5.5 includes two API changes that can turn previously valid requests into HTTP 400 errors:»)
- The model chooses how much reasoning to use for each request, while requests that disable reasoning or set a fixed budget are rejected. (confirmed by the publication itself: evidence; «Thinking is always adaptive. Requests that disable thinking or set a fixed thinking budget are rejected. The model decides how much to think for each request.»)
- Forced tool use is no longer available; Vercel advises asking the model to use the tool and retrying missed calls in client code. (confirmed by the publication itself: evidence; «Forced tool use is retired. Requests cannot require a tool call or force a specific tool. Prompt the model toward the tool, then catch and retry misses in your harness.»)
Publications:
Primary sources:
- https://vercel.com/changelog/claude-opus-5-5-now-available-on-ai-gateway
- https://metr.org/blog/2026-09-22-claude-opus-5-5
- https://platform.claude.com/docs/en/build-with-claude/preserved-thinking
score 86.8 out of 100 · kind: release · update 4