Anthropic's Claude Fable 5.1 Launches on Amazon Bedrock and Claude Platform on AWS

Anthropic's Claude Fable 5.1 is now available on Amazon Bedrock and Claude Platform on AWS, according to an AWS Machine Learning blog post. The model is positioned as an incremental but meaningful upgrade over Claude Fable 5, with Anthropic claiming clear gains on the hardest reasoning benchmarks it tracks internally, including competition mathematics, graduate-level engineering and science questions, and long multi-step problems.

No specific benchmark scores, parameter counts, context window size, or pricing were disclosed in the announcement. AWS points customers to the standard Amazon Bedrock pricing page for cost details rather than publishing numbers directly.

What's new, according to Anthropic

Anthropic describes several practical improvements in Fable 5.1:

• Agentic coding: Handles more of a coding project autonomously, including codebase-spanning features, code review, and performance work across multi-hour sessions. Anthropic claims the model is more honest about failure — it reports when it's stuck rather than disabling a failing test to force a pass.
• Autonomous operation: Built for multi-hour jobs spanning multiple applications, with the ability to plan, invoke tools, recover from failed steps, and provide unprompted status updates.
• End-to-end knowledge work: Can move from an initial question through research, spreadsheet construction, and document drafting, with self-checking along the way. Anthropic frames this as targeting finance, accounting, and healthcare workflows.
• Scientific research: Supports research campaigns spanning literature review, hypothesis generation, modeling, experimentation, and formal verification.
• Usability: Improved clarity in writing and tighter instruction-following, plus more frequent progress updates during long-running tasks.

These are Anthropic's own characterizations; AWS's post does not include independent benchmark comparisons against Fable 5 or competing models.

Covered Model status and data handling

Anthropic has classified Fable 5.1 as a Covered Model — a tier reserved for its most capable systems that comes with additional data retention, safety review, and access requirements. On Amazon Bedrock, using Fable 5.1 requires the aws_review mode, under which AWS retains prompts and outputs for up to 30 days for human safety review conducted within the AWS boundary. Anthropic does not receive this data directly.

Enterprise Frontier Safeguards

Alongside the model launch, AWS and Anthropic introduced Enterprise Frontier Safeguards (EFS), a program letting eligible enterprise customers run Fable 5 and Fable 5.1 with zero data retention (ZDR) on Amazon Bedrock and Claude Platform on AWS. This option is available for internal use through December 31, 2026. AWS says additional safeguards — including keeping prompts and outputs inside a customer's own AWS account under their encryption keys, access policies, and audit logs, with automated rather than human safety review — will roll out later in the year.

Availability and access

Fable 5.1 is accessible today through Amazon Bedrock's US Geo CRIS (us.) and Global CRIS (global.) inference profiles, and through both bedrock-runtime and bedrock-mantle endpoints in AWS GovCloud (US). It's also available via Claude Platform on AWS in North America. Developers can invoke it through the Anthropic Messages API, or through Bedrock's Invoke and Converse APIs via the AWS CLI and SDKs. As a reasoning model, Fable 5.1 can return a thinking block ahead of its text response, so integrations need to select the text block explicitly rather than assume a fixed response index.

What this means

The Covered Model designation and the new Enterprise Frontier Safeguards signal that Anthropic and AWS are treating Fable 5.1 as a genuinely frontier-capability release, not a routine point update — the added review requirements and data controls exist precisely because the model can operate with more autonomy over longer horizons. For regulated industries like finance and healthcare, the ZDR option through 2026 is likely the more consequential news than the capability claims themselves, since it addresses a standing objection to deploying frontier models on sensitive data. Until Anthropic publishes concrete benchmark numbers and AWS discloses per-token pricing, buyers evaluating Fable 5.1 against Fable 5 or rival models are working from qualitative claims only.