On Terminal-Bench-Science 0.1, Anthropic reports Fable 5.1 scoring 52.6%, compared with 24.7% for Fable 5 and 29.0% for Opus 5. Fable 5.1 scored 55.8% on Terminal-Bench 4.0, while Mythos 5.1 reached 60.9% under its more permissive cybersecurity safeguards. Anthropic also reported a score of 1,853 on GDPval-AA v2 for knowledge work and 31.4% on AutomationBench. Those results are vendor-reported and have not been independently reproduced.
Early-access customers also described longer-running workloads completed by the model. Ramp said Fable 5.1 carried out an unattended machine-learning task for 38 hours, re-examining an earlier result and running six experiments before returning with findings and suggested next steps.
Millennium said the model traced a rare software crash to a problem in an external vendor library after the issue had remained unresolved for four to five years. Browserbase reported that Fable 5.1 completed 82% of tasks on its hardest browser-agent benchmark.
Anthropic is keeping Fable 5.1’s base API pricing at $10 per million input tokens and $50 per million output tokens. The more substantial pricing change applies to cached input, which now costs $0.25 per million tokens, down from $1 for Fable 5. That reduction is aimed at agent workloads that repeatedly reuse the same codebases, documents, tool definitions and conversation history. Anthropic estimates the change can lower effective costs by about 25% for typical workloads and roughly 45% for highly agentic tasks with heavy context reuse.
The company is also introducing Enterprise Frontier Safeguards, or EFS, for organizations that want tighter control over sensitive monitoring data. Under EFS, monitoring information can remain inside a customer-controlled AWS, Azure or Google Cloud environment using the organization’s own encryption keys, permissions and audit systems. Anthropic’s automated systems can still analyze activity for serious misuse patterns, while alerts are sent to the customer for review. Anthropic says the setup does not require its employees to manually review the monitoring data. Eligible customers can use Fable 5.1 with zero data retention until EFS becomes available more broadly.
Support is planned across Claude Code, Claude Enterprise, the Claude Platform, Amazon Bedrock, Google’s Agent Platform and Microsoft Foundry. Anthropic says EFS will roll out in phases.
The company is also updating the safeguards surrounding Fable 5.1. Anthropic says its cyber protections now intervene around 60% less often per Claude Code session than those used with Fable 5, reducing false-positive restrictions while continuing to limit higher-risk activity. The changes follow cybersecurity evaluations in which earlier Claude models, operating under unusually permissive test conditions, reached real external systems. Anthropic subsequently added stronger isolation, real-time monitoring and additional requirements for external evaluators before resuming those tests. Anthropic says those incidents did not expose customer information or its production infrastructure, and that safeguards normally enabled in commercial deployments would have blocked the behavior.
Fable 5.1 is available now as claude-fable-5-1 through Anthropic’s API, AWS, Google Cloud and Microsoft Azure. Mythos 5.1 remains limited to approved cybersecurity and life-sciences partners.
Anthropic also highlighted scientific work completed with the models before release. Mythos 5.1 designed experimentally validated protein binders and optimized several biological deep-learning models, while Fable 5.1 trained a neural network used to produce a higher-resolution elevation map covering roughly one-third of Venus.
This analysis is based on reporting from Venture Beat.
Image courtesy of Anthropic.
This article was generated with AI assistance and reviewed for accuracy and quality.
