Anthropic is pushing its Claude model lineup further into long-running coding, research, and knowledge-work tasks with the launch of Claude Fable 5.1 and Claude Mythos 5.1.

Announced on September 1, the two models are built on the same underlying model but are offered with different safeguard configurations. Claude Fable 5.1 is the generally available version, while Claude Mythos 5.1 is reserved for vetted organizations working on advanced cybersecurity and life sciences research.

Anthropic describes Fable 5.1 as its most capable generally available model for coding and knowledge work, with a particular focus on tasks that can run for hours or span multiple stages. The company says the model is designed to plan work, use tools, recover from failures, and continue working with limited supervision.

Claude Fable 5.1 targets long-running AI work

Fable 5.1 is aimed at workloads that go well beyond answering individual prompts. Anthropic says the model can work across large codebases, conduct research, operate browsers and other tools, and handle multi-stage enterprise projects.

In Claude Code, Fable 5.1 defaults to Anthropic’s High effort setting, while Claude Cowork and Claude.ai default to Medium effort. The company says lower effort levels can deliver performance comparable to or better than Fable 5 while reducing the amount of compute required.

The model also brings stronger computer-use and vision capabilities. Anthropic says Fable 5.1 can interpret diagrams, charts and tables inside files and PDFs, while using vision to inspect the results of its own coding work and compare them against the intended design.

That makes the model particularly relevant to workflows where the AI is expected to do more than generate text or code. Instead, it can research a problem, make changes, test the result, inspect its output and continue iterating.

Anthropic points to several early users reporting improvements on complex projects. Millennium, for example, said Fable 5.1 identified the cause of a rare software crash that had remained unexplained for several years. MongoDB described using the model to research an internal codebase and documentation before allowing it to work unattended on a complex prototype.

Benchmark gains over Fable 5

Anthropic’s published benchmarks show substantial gains over the previous Fable 5 model.

On Terminal-Bench-Science 0.1, which measures agentic scientific research, Fable 5.1 scored 52.6%, compared with 24.7% for Fable 5. On Terminal-Bench 4.0 for agentic coding, Fable 5.1 scored 55.8%, while Mythos 5.1 reached 60.9% in the configuration shown by Anthropic. Fable 5 scored 42.0%.

The model also scored 1,853 on GDPval-AA v2 for knowledge work, compared with 1,723 for Fable 5 and 1,824 for Anthropic’s Opus 5.

On Humanity’s Last Exam, Fable 5.1 scored 60.9% without tools and 65.0% with tools. Fable 5 recorded 57.8% and 63.8%, respectively.

For agentic coding, Fable 5.1 reached 73.4% on CursorBench 3.2.0, compared with 70.5% for Fable 5. Anthropic also reports gains on AutomationBench and OSWorld, although it notes that its production safeguards can affect results on some of these evaluations.

These figures are Anthropic’s own evaluations, so they should be read in the context of the company’s testing methodology and the safeguards enabled during testing.

Fable 5.1 and Mythos 5.1 are the same model

One of the more interesting parts of the launch is that Fable 5.1 and Mythos 5.1 aren’t separate model architectures.

Anthropic says they are the same underlying model with different levels of safeguards.

Fable 5.1 is designed for general availability, with safeguards that restrict certain high-risk cybersecurity and biology capabilities. Mythos 5.1 removes or reduces some of those restrictions for vetted researchers working on defensive cybersecurity and life sciences applications.

The distinction is particularly important for cybersecurity. Fable 5.1 can now identify software vulnerabilities in source code, according to Anthropic, but it remains restricted from developing exploits. Anthropic says its updated cybersecurity safeguards produce 60% fewer false positives than its previous system.

For biology, Mythos 5.1 is available through a Life Sciences Verification Program, while Anthropic says it is working with the U.S. government on access to the model’s advanced biology capabilities. The company plans to expand enrollment to additional scientists.

Mythos 5.1 access remains limited to a small group of vetted organizations. Anthropic says the model is currently available to qualifying cyberdefenders and life scientists through its trusted-access programs, with broader availability dependent on the company’s safety and access framework.

Anthropic is also cutting the cost

Fable 5.1 isn’t simply a performance upgrade. Anthropic is also changing the economics of using the model.

The company says Fable 5.1 will cost about 25% less than Fable 5 for typical workloads when usage is billed by token. For highly agentic workloads, Anthropic estimates savings of up to around 45%.

The biggest change is cache-read pricing. Fable 5.1 cache reads cost $0.25 per million tokens, a 75% reduction from Fable 5. Standard pricing is $10 per million input tokens and $50 per million output tokens.

That matters for agentic systems because an AI agent can repeatedly reuse large amounts of context while working through a long task. Lower cache-read costs can therefore have a much larger effect on the final cost than simply reducing the price of individual prompts.

Anthropic says the model is available through its own platform as well as Amazon Web Services, Google Cloud and Microsoft Azure. Developers can access it through the Claude API using the claude-fable-5-1 model identifier.

A different approach to AI safeguards

Anthropic is also introducing a new approach to enterprise data handling alongside Fable 5.1.

Its Enterprise Frontier Safeguards system is designed to let enterprise customers keep data inside cloud infrastructure controlled by the customer rather than Anthropic while maintaining safeguards against adversarial use. Anthropic says the system will be introduced to enterprise customers in phases beginning later this fall.

Until that system becomes available, eligible customers can use Fable 5.1 with zero data retention. Mythos 5.1, meanwhile, currently operates under a 30-day data-retention policy for safety monitoring.

The launch therefore represents more than another incremental Claude update. Anthropic is positioning Fable 5.1 around a future in which AI systems increasingly operate as long-running agents rather than simple conversational assistants, while Mythos 5.1 provides a controlled path to higher-risk capabilities for specialized scientific and cybersecurity work.

For everyday Claude users, Fable 5.1 is the version that matters most. For developers and enterprises building autonomous workflows, however, the combination of higher performance, lower cache costs and longer-running agent capabilities could make the model considerably more practical for workloads that previously required substantial human supervision.