Claude Fable 5.1 is now available in Microsoft Foundry

Fable 5.1 and Mythos 5.1 are the same underlying model. Claude Fable 5.1 is a public preview version with safeguards that fallback for cyber and bio capability queries; Claude Mythos 5.1 is Anthropic’s most capable model for cybersecurity and biology research, keeping its full cyber and bio capabilities as Claude Mythos 5 did, with access limited. 

Claude Fable 5.1 delivers frontier intelligence for ambitious tasks across coding, scientific research, and enterprise workflows. While the model delivers improvements across traditional reasoning benchmarks, the more significant advancement is how effectively it can sustain performance across longer, more complex workflows, planning multi-step tasks, using tools, adapting as conditions change, and carrying work through to completion. For organizations moving from AI assistants to AI systems that can take on meaningful units of work, that difference matters. 

Built for work that extends beyond a single prompt 

Many enterprise workloads cannot be completed in one interaction. A software engineering task may span dozens of applications. A performance investigation may require multiple rounds of analysis and testing. An agent may need to navigate applications, process a backlog, coordinate tools, respond to new information, and continue working over an extended period. 

These scenarios require more than strong point-in-time reasoning. They require a model that can maintain context, make decisions across multiple steps, recover when an approach fails, and continue progressing toward an objective. 

Claude Fable 5.1 delivers significant improvements in these long-horizon workflows. For developers, this expands the types of engineering tasks that can be delegated to AI from code generation and debugging to larger features, codebase-level changes, code review, and performance optimization. 

For agentic applications, it means greater reliability across workflows that involve browsers, tools, enterprise systems, and managed agents operating over longer periods. 

When evaluating Claude Fable 5.1, consider starting with the workloads that have historically been hardest to automate: the ones that require sustained reasoning, multiple tool calls, and decisions that build on one another. 

More capable reasoning and more reliable execution 

Claude Fable 5.1 advances reasoning performance across complex mathematics, graduate-level science and engineering, and multi-step problems where an incorrect assumption early in the process can affect the entire result. But higher benchmark scores are only part of what determines whether a model is ready for production. 

Equally important is how reliably the model behaves when a task does not go according to plan. Claude Fable 5.1 is better at recognizing when it is blocked, communicating incomplete work, and avoiding shortcuts that create the appearance of success without actually solving the problem. 

In software engineering, for example, a model should not resolve a failing test by simply disabling the test. In an analytical workflow, it should not fill gaps in the data with unsupported assumptions and present the result as complete. 

Claude Fable 5.1 shows up as better judgment on ambiguous tasks and fewer confident wrong answers. If it gets stuck it says so instead of reporting success, and it is less likely to take shortcuts like disabling a failing test. It plans the work, uses the tools it needs, recovers when a step fails, and keeps you posted without being asked. 

Taking knowledge work from analysis to output 

Claude Fable 5.1 extends these capabilities beyond software development. Enterprise knowledge work is rarely a single question followed by a single answer. A typical workflow may involve researching a topic, analyzing information, performing calculations, producing an artifact, reviewing the result, and refining it for a particular audience. Fable 5.1 is designed to work across more of that end-to-end process. 

For example, a model can help research a business problem, analyze supporting information, work with spreadsheets, develop a recommendation, and turn the results into a memo or presentation while checking its work along the way. 

These capabilities are particularly relevant for knowledge-intensive functions such as finance, accounting, healthcare, and professional services, where the value of AI increasingly depends on completing workflows, not simply generating content. Research organizations can apply the same capabilities across activities such as literature review, hypothesis generation, experimental design, results interpretation, and longer-running theoretical work, including workflows that use formal verification tools. 

The opportunity is to move from asking where AI can assist within a workflow to determining which workflows AI can help execute more completely from beginning to end. 

Bringing frontier intelligence into production with Microsoft Foundry 

Access to a capable model is only one part of building a production AI system. 

Organizations also need to evaluate models against their own data and workloads, connect them to tools and enterprise knowledge, build and orchestrate agents, observe how systems behave in production, and apply the security and governance controls required by the business. Microsoft Foundry brings these capabilities together in one platform. 

Developers can evaluate Claude Fable 5.1 alongside other models, select the right model for each workload, integrate models into agents and applications, and operate those systems as their requirements evolve. This flexibility becomes increasingly important as AI applications grow more sophisticated. The goal is not to choose a single model for every scenario, but to build an adaptable AI system that can apply the right intelligence to each workload while maintaining consistent operational controls. 

Migrating from Claude Fable 5 

For teams already using Claude Fable 5, existing prompts should generally carry forward to Fable 5.1. However, migration is also a good opportunity to revisit prompts that were designed specifically to compensate for previous model behavior. Instructions such as repeated warnings not to guess, multiple layers of self-verification, or highly prescriptive completion criteria may no longer be necessary in every case. Excessive scaffolding can sometimes limit the effectiveness of a more capable model. 

Rather than carrying every existing instruction forward unchanged, evaluate Claude Fable 5.1 against representative production workloads. Compare quality, reliability, latency, and the effectiveness of your existing verification mechanisms, then simplify where the evidence supports it. 

Enterprise Frontier Safeguards from Anthropic 

Anthropic also announced Enterprise Frontier Safeguards, a solution that gives customers the option to safely deploy Anthropic’s most capable models while storing their data in cloud infrastructure they control. 

Get started with Claude Fable 5.1 in Microsoft Foundry 

Claude Fable 5.1 is available in the Microsoft Foundry model catalog today. 

As you evaluate it, go beyond simple prompt comparisons. Test it against the work that is difficult precisely because it requires sustained reasoning: a feature that spans an entire codebase, an agent that must operate across tools and systems, an analysis that takes multiple steps to complete, or a research workflow that requires the model to continuously evaluate and refine its approach.  Try it today @ AI Model Catalog | Microsoft Foundry Models.