This paper introduces Claude Fable 5.1 and Claude Mythos 5.1, the latest iteration of Anthropic’s models optimized for coding, knowledge work, and complex, agentic scientific research. While Fable 5.1 and Mythos 5.1 share the same underlying architecture, they are differentiated by their safety profiles: Fable 5.1 is the generally available version, whereas Mythos 5.1 is restricted to trusted access programs, specifically tailored for secure research in life sciences and cybersecurity.

Core Methodology and Performance

Fable 5.1 demonstrates a significant leap in performance over its predecessor, Fable 5, particularly in long-running agentic tasks. The methodology for these improvements focuses on architectural enhancements that allow for deeper, more reliable problem-solving and the ability to address the root causes of software errors rather than relying on superficial fixes.

Empirical benchmarks highlight these advancements:

• Agentic Scientific Research: On the *Terminal-Bench-Science 0.1* metric, Fable 5.1 achieved a score of $52.6\%$, compared to $24.7\%$ for Fable 5.
• Agentic Coding: On *Terminal-Bench 4.0*, Fable 5.1 (via the Mythos configuration) reached $60.9\%$, outperforming the baseline Fable 5.
• Multidisciplinary Reasoning: Using *Humanity’s Last Exam*, Fable 5.1 reached a $65.0\%$ pass rate with tool usage, showing improvement over Fable 5’s $63.8\%$.

The performance gap between Mythos 5.1 and Fable 5.1 is notably narrower than in previous versions, attributed to enhanced, less restrictive cybersecurity safeguards that reduce false-positive interventions while maintaining security integrity.

Pricing and Data Privacy

Anthropic has optimized the cost structure for Fable 5.1 by reducing pricing on "cache reads"—the process where the model references pre-processed and stored inputs. This adjustment results in:

• An approximate $25\%$ reduction in costs for typical workloads.
• Up to a $45\%$ reduction for highly agentic workloads.

Regarding privacy, the paper introduces *Enterprise Frontier Safeguards (EFS)*. This system allows enterprise customers to maintain total control over their data within their own cloud infrastructure, effectively achieving a zero-data-retention environment while preserving the model’s adversarial defense capabilities.

Refined Safeguards

The paper details a significant evolution in safety protocols:

• Cybersecurity: Safeguards are now $60\%$ more precise, reducing the incidence of false positives. The model is now permitted to assist in identifying software vulnerabilities, though it remains restricted from generating actionable exploits.
• Biology: A collaborative partnership with the US government establishes a structured access program for the Mythos 5.1 model, ensuring that its advanced capabilities in the life sciences are deployed exclusively within verified, secure research contexts.

In summary, Fable 5.1 and Mythos 5.1 represent a shift toward high-effort, agentic computing that prioritizes cost efficiency, institutional-grade data privacy, and improved nuance in safety interventions.