Anthropic unveiled its latest flagship artificial intelligence (AI) model line, Claude Fable 5.1 and Claude Mythos 5.1, targeting enterprise demand for complex, long-running agentic workloads while restructuring its pricing to lower operational costs.
The dual-model release relies on identical underlying architecture tailored for different deployment environments. Fable 5.1 serves as the generally available release equipped with standard production safeguards. Mythos 5.1 is restricted to vetted cybersecurity and life-sciences organizations operating within specialized trust programs, granting permissive access to higher-tier research capabilities.
Anthropic is positioning the 5.1 release as a specialized engine for extended problem-solving rather than single-prompt task completion. According to vendor-reported benchmarks, Fable 5.1 achieved a score of 53% on the Terminal-Bench-Science 0.1 evaluation for agentic scientific research—more than double the 25% scored by its predecessor, Fable 5, and surpass OpenAI’s GPT-5.6 Sol (22%).
On software development benchmarks, Fable 5.1 reached 56% on Terminal-Bench 4.0, while unrestricted Mythos 5.1 achieved 61%. The model also demonstrated gains across broader business metrics, posting 31% on AutomationBench and reaching 73% on CursorBench 3.2.0.
Early enterprise partners highlighted the model’s ability to execute multi-hour autonomous tasks.
Investment firm Millennium reported that Fable 5.1 identified the root cause of an obscure software crash inside a third-party library that had eluded engineers for nearly five years.
Corporate finance platform Ramp documented an unattended 38-hour machine learning run where the model autonomously launched six comparative experiments and synthesized actionable findings.
Web automation developer Browserbase cited an 82% task completion rate on its hardest browser-agent evaluation suite, up from 57% with Fable 5.
While Anthropic maintained its base pricing of $10 per million input tokens and $50 per million output tokens for Fable 5.1, it slashed the cost of cached input tokens by 75%, dropping the rate from $1.00 down to $0.25 per million. The aggressive cut directly addresses the economics of persistent AI agents, which frequently re-read extensive system instructions, repositories, and historical logs. Anthropic projects the revised cache rates will reduce effective operational costs by 25% to 45% for highly agentic workflows.
The structural shift follows market data suggesting enterprise hesitation around persistent model costs. Financial data from Ramp showed Fable 5 accounted for just 11% of corporate Anthropic spend two months post-launch, as companies favored lower-cost alternatives like Opus 5.
The deployment arrives alongside enhanced governance features. Anthropic introduced Enterprise Frontier Safeguards (EFS), allowing clients to retain security monitoring data entirely within customer-controlled infrastructure.
The update also integrates invisible text watermarking—supported by a private-preview detection API designed to satisfy European Union compliance standards—and cuts cybersecurity false positives in Claude Code by roughly 60%.