Vercel AI Gateway Adds Support for Claude Opus 5.5, Significantly Boosting Agentic Coding Performance

Vercel AI Gateway has started supporting Anthropic's latest model, Claude Opus 5.5. This model demonstrated the most significant performance improvements over Opus 5 in agentic coding, long-running agent tasks, and knowledge work.

Key Features and API Changes

Opus 5.5 features improved collaboration capabilities in long workflows, reporting in plain text on tasks performed, findings, and requirements for the next steps. Additionally, two important API changes have been applied, which may cause existing requests to return HTTP 400 errors.

• Adaptive Thinking: Requests that disable Thinking or set a fixed budget are rejected. The model determines the amount of Thinking per request, which must be controlled via the effort parameter and prompts.
• Mandatory Tool Use Removed: Specific Tool calls can no longer be forced. You must implement logic to guide via prompts and retry on failure; if JSON output is required, use Structured Outputs.

Technical Specifications and Usage

Claude Opus 5.5 supports a 1M token context window and returns up to 128K tokens. The knowledge cutoff is June 2026. It can be called via the AI SDK, OpenAI-compatible Chat Completions API, and Anthropic Messages API using the model name anthropic/claude-opus-5.5. Fast mode can also be enabled via the anthropic/claude-opus-5.5-fast option. Regional Inference and Zero Data Retention (ZDR) are opt-in, allowing you to pin inference to specific regions (e.g., US) and apply data retention policies upon request. It can be connected to coding agents like Claude Code via the Vercel CLI.