Claude Fable 5.1 Arrives on Amazon Bedrock With New Data Controls

AWS has made Claude Fable 5.1 available through Amazon Bedrock and Claude Platform on AWS, positioning the Anthropic model for demanding work in software development, scientific research, and enterprise operations. The release combines claimed improvements in reasoning and autonomous execution with unusually important deployment conditions: Fable 5.1 is classified as a Covered Model, so its standard use carries additional retention, safety-review, and access requirements.

Those governance details are central to the launch. Under the standard Amazon Bedrock configuration described by AWS, customers must enable a specific review mode before invoking Fable 5.1, and prompts and outputs may be retained for human safety review. A separate Enterprise Frontier Safeguards program offers eligible customers a zero-data-retention route for internal use, although that option has eligibility and time limitations.

The model can be tested in the Bedrock console or invoked through several programmatic interfaces. AWS also provides a Boto3 example that illustrates the required model identifier, request format, token setting, and response-handling logic for a reasoning model.

What AWS and Anthropic claim has improved

According to AWS’s account of Anthropic’s testing, Claude Fable 5.1 improves on Fable 5 on some of Anthropic’s most difficult reasoning evaluations. The cited test areas include competition mathematics, graduate-level engineering and science questions, and lengthy, multi-stage problems in which reliability matters. The source does not provide benchmark names, scores, sample sizes, or independent evaluation results, so the improvement should be understood as a vendor-reported claim rather than a quantified comparison.

AWS describes the practical improvements across five areas:

• Agentic coding: Fable 5.1 is presented as capable of taking on more of a software project, including features that span a codebase, code review, and performance work over multi-hour sessions. Anthropic also claims that the model is more likely to acknowledge when it is stuck and less likely to bypass a problem by disabling a failing test.
• Autonomous operation: The model is designed for jobs that run for multiple hours and involve several applications. AWS says it can plan the work, select tools, recover from failed steps, and provide progress updates without waiting to be prompted.
• End-to-end knowledge work: The announced workflow extends from an initial question through research, spreadsheet construction, preparation of a memo or presentation, and numerical checking. AWS identifies finance, accounting, and healthcare as intended everyday use areas.
• Scientific research: The model is described as supporting research campaigns that move from literature review and hypothesis development to models, experiments, and formal verification.
• Usability: Anthropic reports clearer writing, closer instruction following, and more proactive updates during long tasks.

These descriptions indicate the intended scope of the release, but they do not establish that the model will perform equally well in every codebase, application environment, scientific discipline, or regulated workflow. Organizations evaluating it should distinguish the reported capabilities from performance validated on their own tasks.

Covered Model status changes the retention rules

Anthropic has designated Claude Fable 5.1 a Covered Model. In the source, that designation means the model is subject to additional policies concerning retention, safety review, and access wherever it is offered.

For standard use on Amazon Bedrock, data may be retained for up to 30 days and reviewed by Amazon personnel. Customers configure retention through a mode applied to an AWS account or an Amazon Bedrock project using the data retention API. Fable 5.1 specifically requires aws_review mode to be set before the model is invoked.

In that mode, AWS retains the customer’s prompts and the model’s outputs inside the AWS boundary for human safety review. AWS says using Fable 5.1 in this configuration does not require the data to be shared with the model provider. That distinction does not eliminate retention or human access: it identifies AWS, rather than Anthropic, as the party conducting the described review.

Teams handling confidential, personal, regulated, or otherwise sensitive information therefore need to assess the retention period and human-review condition before adopting the standard configuration. The source points customers to the Amazon Bedrock data-retention and abuse-detection documentation for the detailed policies.

Enterprise Frontier Safeguards offer an eligible-customer exception

Enterprise Frontier Safeguards, or EFS, is a program built jointly by AWS and Anthropic. It is intended to let eligible organizations use Claude Fable 5 and Claude Fable 5.1 while keeping data within cloud infrastructure they control.

Eligible EFS customers can use both models with zero data retention on Amazon Bedrock and Claude Platform on AWS. The announced zero-data-retention availability is limited to internal use through December 31, 2026. The source does not spell out the eligibility criteria, so customers cannot assume that the option applies automatically to every Bedrock account or workload.

AWS also says additional safeguards will become available later in the year. Under that future arrangement, customers will be able to keep prompts and outputs in their own AWS accounts, governed by their own encryption keys, access policies, and audit logging. Safety monitoring is described as automated, with no human review required. Because these additional controls are presented as forthcoming, organizations should not treat them as generally available until their availability and account eligibility are confirmed.

How to start in the Amazon Bedrock console

The shortest path to testing Fable 5.1 is through the Amazon Bedrock console. AWS instructs users to open the console, navigate to Test > Playground, choose Fable 5.1, and submit a prompt directly. The required retention mode still needs to be configured before invocation.

Developers have several programmatic paths. They can use the Anthropic Messages API against bedrock-runtime through the Anthropic SDK. Alternatively, they can continue using the Invoke or Converse APIs on bedrock-runtime through the AWS Command Line Interface or an AWS SDK.

Prerequisites

The AWS example identifies five setup requirements:

1. An active AWS account with access to Amazon Bedrock.
2. The AWS CLI installed and configured.
3. Python 3.10 or later.
4. Boto3, installable with pip install boto3.
5. IAM permissions for bedrock:InvokeModel and bedrock:InvokeModelWithResponseStream.

Important details in the Boto3 request

The supplied Python example creates a bedrock-runtime client in us-east-1 and invokes the model with the identifier global.anthropic.claude-fable-5-1. It sends and accepts application/json, declares the Anthropic API version as bedrock-2023-05-31, and sets max_tokens to 4096.

The sample prompt asks the model to estimate monthly egress expense for an S3 bucket serving 40 TB per month at $0.09 per GB, recommend one architectural change, show its calculation, and keep the response below 120 words. Those figures are part of the demonstration prompt, not published pricing guidance or a reported benchmark result.

The response-processing step contains a useful implementation detail. Because Fable 5.1 is a reasoning model, its returned content may contain a thinking block before the text block. The example therefore searches the content array for a block whose type is text rather than assuming that the user-facing answer always appears at a fixed array index. AWS also links to a Getting Started notebook for further examples.

Availability across AWS environments

At launch, Claude Fable 5.1 is available on Amazon Bedrock through the US Geo CRIS profile, identified by the us. prefix, and the Global CRIS profile, identified by global.. In AWS GovCloud (US), it is offered on both the bedrock-runtime and bedrock-mantle endpoints.

The model is also available through Claude Platform on AWS in North America. AWS directs customers to the Bedrock documentation for the complete list of supported Regions and to the Bedrock pricing page for current commercial details; the announcement itself does not provide model pricing.

What prospective users should take away

Claude Fable 5.1’s AWS launch brings together a model intended for long-running, tool-using work and a deployment policy designed for a high-capability Covered Model. The claimed gains span coding, research, knowledge work, autonomous operation, and general usability, but the source supplies no numerical benchmark table with which to measure those gains.

The immediate implementation decision is therefore not only which API to use. Customers must also decide whether the standard aws_review arrangement—with retention for up to 30 days and human review by Amazon personnel—is suitable for their data. Eligible organizations can consider EFS zero data retention for internal use through December 31, 2026, while the additional customer-controlled safeguards remain described as a later release.

About the source authors

The AWS post credits Dani Mitchell, Aamna Najmi, Sofian Hamiti, Antonio Rodriguez, and Ayan Ray. Their stated backgrounds cover Amazon Bedrock Knowledge Bases, governance and operationalization of generative AI, AI solution leadership, generative AI technology, partner architecture, and Anthropic adoption on AWS. Sofian Hamiti is described as having more than 12 years of experience building AI solutions and leading teams.

Source: This article synthesizes the AWS Machine Learning Blog announcement, “Introducing Claude Fable 5.1 on AWS.”

Definition. Claude Fable 5.1 is an Anthropic Covered Model offered through Amazon Bedrock and Claude Platform on AWS for reasoning-intensive, long-running work.