Intel / article

Claude Opus 5.5 now available on AI Gateway

Source: vercel.com
Author: Rohan Taneja
Date: Published Sep 22, 2026

Context

Moving an existing application to Anthropic's Claude Opus 5.5 through Vercel AI Gateway can require changes to request settings. The integration announcement identifies two changes that turn previously valid requests into HTTP 400 errors: disabling thinking or giving it a fixed budget, and requiring the model to call a tool.

Thinking is always adaptive: the model decides how much to think on each request, and developers steer it with the effort setting and prompting. Forced is retired, so an application must prompt the model toward a tool and catch and retry the occasions when it does not make the call. Applications that previously forced a tool call to return JSON are directed to structured outputs. These changes affect the surrounding application logic, so migration involves more than replacing the model name.

Key takeaways · AI-distilled

Two Opus 5.5 API changes can turn previously valid requests into HTTP 400 errors: thinking is now always adaptive (fixed thinking budgets or disabling thinking are rejected), and forced tool-call requirements are retired entirely.
Anthropic states Opus 5.5 matches Fable 5.1-level performance while running about 30% faster and about 40% cheaper per task than Opus 5, per Vercel's changelog.
Opus 5.5 has a 1M-token , returns up to 128K output tokens, and carries a June 2026 .
AI Gateway offers opt-in Regional (pinning requests to a geographic zone) and Zero Data Retention controls, and passes through provider pricing with no platform fee, including on Bring-Your-Own-Key requests.

Terms in this piece · Glossary

tool use — A model's ability to call external functions — run code, search the web, edit files — instead of only generating text.
context window — The maximum amount of text a model can consider at once — its working memory for the current conversation or task.
knowledge cutoff — The date after which a model saw no training data — everything later has to reach it through search, tools, or the context window.
inference — Running a trained model to get answers — the phase where AI is actually used, as opposed to trained.

Read the source vercel.com