Vercel AI Gateway begins supporting Anthropic Claude Fable 5.1

Vercel AI Gateway has started supporting Anthropic's latest model, Claude Fable 5.1. This model offers improved performance in long and complex workflows, such as agentic coding, knowledge work, and multi-step research.

Safety Classifiers and Fallback Features

Fable 5.1 has cybersecurity and biological safety classifiers enabled by default. While discovering vulnerabilities in source code is allowed, some common coding and debugging requests may be rejected. To handle cases where the safety classifier blocks a request, you can use the model fallbacks feature.

By adding a models array to providerOptions.gateway, the AI Gateway first sends the request to Fable 5.1. If Anthropic rejects the request, it switches to other models (e.g., Claude Opus 5, Claude Sonnet 5) in the order specified in the array and returns the response. This option is supported across all AI Gateway API formats, including Chat Completions, Messages, and OpenAI Responses.

Data Retention Policy and Usage

Anthropic does not support Zero Data Retention (ZDR) for Fable 5.1. Prompt and completion data are retained for 30 days but are not used for training Claude.

To use it in coding agents, run the vercel ai-gateway coding-agents setup command to configure connections with tools like Claude Code, Codex, and Cursor, and then select anthropic/claude-fable-5.1 within the agent.