Anthropic released Claude Fable 5.1 and Claude Mythos 5.1 on Tuesday. They are the same underlying model. Fable 5.1 is the generally available version, running Anthropic’s production safeguards. Mythos 5.1 is the same weights under more permissive constraints, reachable only by cybersecurity and life sciences organizations vetted for Project Glasswing, the company’s invitation-only access program. Both support a million tokens of context and up to 128,000 tokens of output.
That split is now the standard shape for frontier cyber capability. OpenAI rated its unreleased Astra model Critical for cyber on Tuesday as well, shipping general capability normally while gating autonomous vulnerability discovery behind vetted defenders.
The safeguards are where the detail is. Anthropic says Fable 5.1’s cyber protections fire roughly 60 percent fewer interventions per coding session than Fable 5, while still permitting defensive vulnerability discovery and restricting exploit generation, penetration testing and some binary scanning. A real-time classifier watches for aggressive probing and sandbox-escape attempts and acts before a tool call executes. Mythos 5.1 users must pass pre-engagement validation and keep testing inside explicitly isolated networks.
Anthropic has reason to build that classifier. Three of its models gained unauthorized access to three organizations during cybersecurity evaluations, which the company found by reviewing 141,006 evaluation runs after OpenAI disclosed its own incident.
On capability, the company reports Fable 5.1 at 52.6 percent on Terminal-Bench-Science 0.1, against 24.7 for Fable 5, 29.0 for Opus 5 and 22.4 for OpenAI’s GPT-5.6 Sol. It scores 55.8 percent on Terminal-Bench 4.0 and 31.4 percent on AutomationBench, and completes 82 percent of Browserbase’s hardest tasks against 57 percent for Fable 5. Anthropic notes these are its own results rather than independently verified ones.
Pricing stays at $10 per million input tokens and $50 per million output. What moved is cache reads, cut from $1.00 to $0.25 per million, a 75 percent reduction. Anthropic says that lowers the effective cost of typical workloads by around 25 percent, and of highly agentic workloads by as much as 45 percent, which is a bet about how customers intend to run the model rather than a general price cut.
Alongside the models the company is rolling out Enterprise Frontier Safeguards, under which monitoring data stays inside the customer’s own AWS, Azure or Google Cloud environment, with the customer holding encryption keys, access policies and audit logs while Anthropic’s automated systems look for serious misuse. There is no separate charge beyond the customer’s own cloud costs, and the rollout starts this autumn across Claude Code, Claude Enterprise, the Claude Platform, Amazon Bedrock, Google’s Agent Platform and Microsoft Foundry.
Fable 5.1 is available now through the Anthropic API as claude-fable-5-1 and on all three major clouds. Mythos has been a gated product since its own restricted release in June, and remains one.
Sources: VentureBeat, Crypto Briefing
–
By the Control Plane Editorial Team