Claude Fable 5.2 on AWS is now available

Amazon Web Services and Anthropic made Claude Fable 5.2 and Claude Mythos 5.2 available on Amazon Bedrock and Claude Platform on AWS, according to AWS Machine Learning Blog's report. Fable 5.2 is Anthropic's flagship model for coding, math, and science. Mythos 5.2 is offered through verification programs for cyber security and life sciences.

This report covers model capabilities, Bedrock runtime configurations, and upgrade paths for teams deploying autonomous coding agents. It provides technical specifications for developers, researchers, and infrastructure teams evaluating Fable 5.2 on AWS. Readers will find exact migration commands, fallback cache mechanics, privacy timelines, and invocation parameters.

01What shipped in Claude Fable 5.2 on AWS

Claude Fable 5.2 is available on Amazon Bedrock through US Geo CRIS and Global CRIS inference profiles, alongside Claude Platform on AWS. The model handles ambiguous, multi-step engineering assignments, verifies its own output, and runs unattended for hours to days. It interprets prompts, iterates against test results, catches errors, and optimizes execution dynamically.

While Fable 5.2 is generally available, Claude Mythos 5.2 is restricted to authorized organizations through Anthropic verification programs. Mythos 5.2 covers cyber security, security engineering, red teaming, penetration testing, and biomolecular design. Primary use cases for Fable 5.2 include multi-repository agentic coding, multi-day subagent workflows, scientific proofs, quantitative forecasting, and 3D simulation code.

02What improved in Claude Fable 5.2 on AWS

Anthropic alignment testing shows Mythos 5.2 hallucinates less, completes tasks more reliably, and surfaces critical details more consistently than Opus 5.5, Sonnet 5.5, and Mythos 5.1. The model also self-reports operational issues and reasoning gaps more honestly during execution. Anthropic expects Claude Fable 5.2 to demonstrate similar improvements across technical workloads.

Developers also gain cost protections through fallback credit tokens when domain guardrails decline a prompt. If a cyber query is declined, client code retries on Opus 4.8; declined biology queries retry on Opus 5. Because the refusal includes a fallback credit token, Bedrock bills the follow-up retry as cache reads rather than charging to write the identical cache twice.

03What you gain from Claude Fable 5.2 on AWS

Anthropic designates Claude Fable 5.2 as a Covered Model, which normally requires up to thirty days of prompt retention and human safety review. On Amazon Bedrock, Amazon personnel conduct that safety review within the AWS boundary under the aws_review data retention mode. Teams configure this parameter on their AWS account or project before invoking the model.

Enterprise Frontier Safeguards, built with AWS, pairs misuse detection mechanisms with the privacy assurances of zero data retention. The safeguards framework is available immediately for eligible production accounts. Under this agreement, qualified customers can run Fable 5.2 with zero data retention through June 30, 2027, ensuring sensitive enterprise source code is never retained.

04How to get Claude Fable 5.2 on AWS

Developers can test Fable 5.2 visually in the Amazon Bedrock console under Test and Playground, or programmatically using Python 3.9 and Boto3. Required AWS IAM permissions include bedrock:InvokeModel and bedrock:InvokeModelWithResponseStream. Upgrading the local Claude Code CLI tool globally allows developers to switch sessions directly to Fable 5.2 or pin the model at launch.

npm install -g @anthropic-ai/claude-code
claude update

/model fable
/model global.anthropic.claude-fable-5-2

claude --model fable
claude config set model global.anthropic.claude-fable-5-2

For existing codebases, the open-source claude-api skill inside Claude Code automates migration to Fable 5.2 in a single guided pass. It updates model references to global.anthropic.claude-fable-5-2, switches manual thinking configurations to adaptive thinking, removes deprecated beta parameters, and configures output_config.effort. Developers can run the migration command directly or request prompt caching improvements conversationally.

/claude-api migrate

05What to watch after Claude Fable 5.2 on AWS

A key infrastructure development to monitor is the upcoming customer-managed keys version of Enterprise Frontier Safeguards. While the current release provides zero data retention, the next version will let enterprises retain data in dedicated S3 buckets protected by customer-owned encryption keys. AWS and Anthropic have confirmed development is underway, though a formal release date is not yet announced.

Organizations operating under strict privacy mandates should track the June 30, 2027 milestone when current zero data retention terms expire. Teams relying on safety reviews must verify aws_review remains active on their accounts so Amazon personnel conduct reviews within AWS boundaries. Additionally, the open-source claude-api skill in CodeRabbit, JetBrains IDEs, Resolve AI, and Warp will receive continuous updates.