Introducing Claude Fable 5.2 on AWS

AWS announced on its Machine Learning Blog that Anthropic's Claude Fable 5.2 and Claude Mythos 5.2 are now available on Amazon Bedrock and on Claude Platform on AWS. Fable 5.2 is described as Anthropic's most capable generally available model for technical work, aimed at coding and the sciences. Mythos 5.2, by contrast, is not generally available: it is offered through Anthropic's verification programs for advanced work in cyber security and life sciences.
According to the announcement, Fable 5.2 is built for hard problems in coding, math and science. It handles ambiguous, multi-step tasks, checks its own output, and can run unattended for hours or even days, iterating on prompts and correcting its own errors as it goes. On the quality side, the post cites Anthropic's alignment testing for Mythos 5.2, which reportedly hallucinates less, completes tasks more reliably, surfaces important information more consistently and self-reports issues more honestly than Opus 5.5, Sonnet 5.5 and Mythos 5.1. Anthropic expects Fable 5.2 to behave similarly, though the post frames that as an expectation rather than a measured result.
The two models are split by risk domain. Fable 5.2 is pitched at agentic coding across large multi-repository codebases, long-running agents coordinating many subagents on multi-day work, scientific research including formal mathematical proofs, financial analysis such as quantitative research and forecasting, and design work for 3D, hardware and motion that generates environments, components and simulations in code. Mythos 5.2 is confined to security engineering, red teaming and penetration testing on one side, and life sciences research, especially biomolecular design and research synthesis, on the other.
Because Fable 5.2 can decline certain requests, the post explains a fallback path. A declined cyber security query can be retried on Opus 4.8, and a declined biology query on Opus 5. On Bedrock this retry is the application's responsibility rather than an automatic switch. If the declined request had hit the prompt cache, the refusal carries a fallback credit token; sending that token with the retry bills the retry as cache reads, so the same cache is not written and paid for twice. This is a detail worth noting for teams building automated routing: handling refusals is part of the integration work, not something the platform does for you.
Data handling carries conditions. Anthropic designates Fable 5.2 a Covered Model, which by default requires retention for up to 30 days and human safety review. On Bedrock, Amazon personnel perform that review inside the AWS boundary, and customers must set the aws_review data retention mode on their account or project before invoking the model. Enterprise Frontier Safeguards (EFS), developed with AWS, is intended to pair zero data retention with misuse detection by keeping data in customer-controlled cloud infrastructure. Until EFS is available, eligible customers can use Fable 5.2 with zero data retention for internal use through at least June 30, 2027. In practice, teams with strict data-residency or no-retention requirements should confirm which mode applies to them before moving workloads over, since the default path involves human review.
Getting started is straightforward. In the Bedrock console, users open Test, then Playground, and select Fable 5.2. Programmatically, the model can be called through the Anthropic Messages API against bedrock-runtime via the Anthropic SDK, or through the Invoke and Converse APIs using the AWS CLI and AWS SDK. The post's Python example requires an AWS account with Bedrock access, the AWS CLI configured, Python 3.9 or later, boto3, and IAM permissions for bedrock:InvokeModel and bedrock:InvokeModelWithResponseStream. The sample uses the model ID global.anthropic.claude-fable-5-2, sets anthropic_version to bedrock-2023-05-31 and max_tokens to 4096. One practical hint stands out: because Fable 5.2 is a reasoning model, responses may include a thinking block ahead of the text block, so code should select the block whose type is text rather than index into a fixed position. Hardcoding an index is a common source of breakage when reasoning models are swapped in.
For migrations, the post points to the claude-api skill in Claude Code, an open-source skill that encapsulates working Claude API patterns, including which agent patterns suit a job, parameter changes between model generations, and when prompt caching applies. Running /claude-api migrate scans a working directory and applies changes in one guided pass, or developers can ask conversationally for an upgrade. The skill swaps model IDs, moves manual thinking settings to adaptive thinking where appropriate, removes deprecated beta headers and parameters, and suggests output_config.effort values for the use case. It also detects the deployment platform and adjusts ID formats accordingly, such as the Bedrock structured form used above. The skill ships in Claude Code and is bundled into CodeRabbit, JetBrains IDEs including Junie, Resolve AI and Warp, and is open source at anthropics/skills on GitHub. This suggests a broader pattern: model migrations increasingly come with tooling that edits code directly, which lowers upgrade friction but means teams should review automated diffs rather than accepting them blindly.
Availability is immediate on Bedrock through US Geo CRIS and Global CRIS inference profiles, with the full regional list in Bedrock documentation and pricing on the Bedrock pricing page. Both models are also available via Claude Platform on AWS in every Region where that service operates.
Why it matters: teams building coding agents, scientific tooling or financial analysis now have a generally available Anthropic frontier model on Bedrock, while security and biology use cases are gated behind verification. The likely practical changes, by inference, are refusal-handling code, a data retention mode decision before first invocation, and possibly a scripted migration via the claude-api skill.
Based on reporting from the original publisher. Visit the source for full context and later updates.
Publisher excerpt
Claude Fable 5.2 and Claude Mythos 5.2 are now available on Amazon Bedrock and Claude Platform on AWS. Learn what's new in both models, why Fable 5.2 matters for your technical workloads, and how to get started building with it on Amazon Bedrock.