Amazon adds Anthropic’s Claude Fable 5 to AWS Bedrock, extending AI agents to long-running tasks and document-heavy work

Amazon Web Services on Tuesday announced that Anthropic’s Claude Fable 5 is now available to customers on AWS, bringing the latest “Mythos-class” capabilities to Amazon Bedrock and to Anthropic’s native Claude Platform on AWS. The move is positioned as a step toward more capable, safer AI agents that can carry out extended work and handle tasks that involve complex plans, ongoing progress checks, and content inside files and documents.

In AWS’s announcement, the company said Claude Fable 5 can perform long-running, asynchronous execution through agent-style workflows such as Claude Code, where the system can keep working for days rather than requiring constant human prompts. AWS described the model as able to plan an approach, compare its progress against a stated goal, and refine its output as it goes, addressing a key limitation that has often forced earlier models to break down complicated assignments into shorter loops.

AWS also highlighted Claude Fable 5’s “advanced vision” abilities, saying it can interpret diagrams, charts, and tables embedded in files and PDFs. It pointed to use cases spanning research and document-heavy work in finance, legal, analytics, architecture, and gaming, and said the model can use vision to critique coding outputs against goals, not just generate them.

Amazon added that Claude Fable 5 is designed with “proactive self-verification,” including the ability to update its own skills based on learnings, develop harnesses and evaluations, and verify its work. While the company did not provide additional technical detail in the release, the framing indicates that the model is intended to assess its own results more directly during multi-step projects.

For enterprise deployments, the availability on Amazon Bedrock is the centerpiece. Bedrock is AWS’s managed service for building and running foundation-model applications within the AWS environment, with options that can help organizations maintain enterprise security and regional data residency requirements. Amazon said customers can access Claude Fable 5 through the Bedrock Runtime console and the new Bedrock Mantle console.

Bedrock Mantle, per Amazon, is designed to support developers with “project-aware live documentation” and auto-prefilled code snippets. The company’s description suggests the workflow is meant to reduce friction when building applications that use the model, especially for teams moving from prototypes to production settings.

Anthropic, in a separate announcement, positioned Claude Fable 5 as a Mythos-class model that it said exceeds the performance of any model it has previously made generally available. It also addressed safety tradeoffs, saying releasing a model this capable can carry risks, particularly around areas such as cybersecurity, and that it launched Fable 5 with safeguards that may redirect some requests to its next-most-capable model, Claude Opus 4.8.

Anthropic said those safeguards are tuned conservatively and “sometimes catch harmless requests,” but that they trigger, on average, in less than 5% of sessions. It also noted that as more capable models arrive, it expects to improve the safeguards to reduce false positives. The Anthropic release did not describe AWS-specific policy behavior, but it frames what customers should expect when queries touch sensitive topics.

Beyond the immediate product launch, Amazon’s decision reflects intensifying competition around developer access to frontier AI models on cloud platforms. AWS has increasingly marketed agentic AI, long-running workflows, and managed model hosting as differentiators, and adding Anthropic’s newest flagship model to Bedrock and the Claude Platform on AWS gives customers another option for building software and analysis tools.

What remains less clear is the scope of availability and performance targets for Claude Fable 5 on AWS. Amazon’s release did not specify throughput limits, pricing, latency, or the size of the “guardrails” policy set customers should expect inside Bedrock. Customers also will likely want to validate how well the model sustains multi-day tasks in their own environments and how it behaves with their specific document formats, toolchains, and evaluation requirements.

Looking ahead, the practical test will be whether developers can reliably integrate Claude Fable 5 into production-grade agent workflows, especially where tasks span days, require tool use, and involve vision-based extraction from PDFs and diagrams. AWS and Anthropic will also be watching how safeguards perform in real deployments, particularly whether the “less than 5%” conservative trigger rate holds broadly across use cases.