Claude Fable 5.1 adds agent controls, but migration and Bedrock constraints matter

Anthropic's platform documentation specifies Claude Fable 5.1 as `claude-fable-5-1`, with matching identifiers for Amazon Bedrock, Google Cloud, Microsoft Foundry, and Claude Platform on AWS. It lists a one-million-token context window, 128,000 maximum output tokens, adaptive thinking, and a high default effort setting. Those are interface and configuration facts rather than independent performance results, but they materially affect how an integration budgets context, latency, and output handling.

The comparison with Fable 5 is mostly economic and workflow-oriented. Anthropic says Fable 5.1 retains the earlier model's input and output rates while cache reads cost a quarter as much, and it positions the model for long-running coding, multistep research, and document or spreadsheet work. Axios independently reported the same advertised price points and cache-read reduction. The cited material does not constitute an independent controlled evaluation, so capability gains remain vendor-reported rather than confirmed benchmark improvements.

The migration details are more operationally important than the capability list. Anthropic warns that forced tool use returns an error, earlier models cannot read Fable 5.1 thinking blocks, and editing earlier turns invalidates those blocks. It also lists beta per-message effort, turn-scoped system messages, and readable progress updates between tool calls. Teams moving an existing agent should regression-test tool selection, conversation edits, and thinking-block handling before treating this as a model-ID-only routing change.

For Bedrock users, the model card adds hard integration constraints: temperature must be 1.0 or unset, top_p must be 0.99 or unset, the two cannot be set together, and top_k is unsupported. The card records support for Messages, Converse, and Invoke on `bedrock-runtime`, while Responses and Chat Completions are unsupported there. Portability work should therefore be treated as an API-compatibility task rather than a simple endpoint switch.

Safety and data handling remain part of the deployment decision. AWS says the model is subject to up to 30 days of retention and human review in its required `aws_review` mode, while eligible Enterprise Frontier Safeguards customers can use a zero-data-retention arrangement for internal use through December 31, 2026. The Bedrock card separately says refusal rates are materially higher than on earlier Claude models and directs clients to handle `stop_reason: "refusal"` as a primary response path. Production agents should test retention-policy eligibility and refusal recovery before relying on unattended workflows.