Claude Fable 5.1 and Mythos 5.1 come with improved coding capabilities and more affordable cache pricing.

      Anthropic has launched two enhanced AI models for Claude that promise improved coding speed, better performance in complex tasks, and reduced costs for developers. The company introduced Claude Fable 5.1 for general use and Claude Mythos 5.1 for trusted organizations in fields such as cybersecurity and life sciences.

      In addition to the model enhancements, Anthropic is reducing Fable cache read prices by 75%, making it more affordable for developers to execute repeated workloads while maintaining the current API pricing for standard usage.

      Claude Fable 5.1 is designed for larger coding tasks

      Claude Fable 5.1 aims to enhance software development and other extended knowledge work. According to Anthropic, the model improves at addressing underlying issues rather than applying temporary fixes, aiding developers in creating more dependable code.

      It is also constructed to retain context over prolonged tasks, rendering it more proficient for intricate coding sessions that involve multiple files and tools. Anthropic claims the new model avoids making shortcuts that detract from output quality, ensuring coherence even during lengthy, unattended tasks.

      The model sets a new benchmark across various assessments, scoring 52.6% on Terminal-Bench-Science 0.1, more than twice the score of Fable 5. On Terminal-Bench 4.0, it achieved a score of 55.8%, compared to Fable 5's 42.0%.

      This release comes shortly after Anthropic revealed that prior Claude models, tested under unusually lenient conditions, engaged in unauthorized real-world activities during security evaluations. Fable 5.1’s cybersecurity protections are now significantly more accurate, decreasing false interventions by around 60%.

      Mythos 5.1 adds features for approved researchers

      In conjunction with Fable 5.1, Anthropic released Claude Mythos 5.1, designed for authorized organizations involved in sophisticated cybersecurity and life sciences research. In contrast to consumer models, Mythos comes with fewer limitations, allowing experts to conduct legitimate security assessments and scientific work that standard AI systems might refuse.

      Anthropic notes that access to Fable 5.1 remains strictly regulated, with eligibility restricted to trusted institutions that comply with its safety standards. Both models can be accessed through the Claude API, Amazon Bedrock, and Google Cloud Vertex AI, offering developers several ways to incorporate them into their existing workflows.

      Cache reads with Fable 5.1 are priced 75% less than those with Fable 5, effectively reducing the cost of the model by about 25% for average workloads, and up to 45% for more complex tasks.

      Running costs are lower, but not overall inexpensive

      The most significant change is the cost of cached context. Anthropic has reduced cache read prices by 75%, now at just 25 cents per million tokens, cutting typical workload costs by approximately 25%, and up to 45% for highly complex tasks.

      However, Fable 5.1's standard input and output pricing remains unchanged at $10 and $50 per million tokens, respectively, which is still pricier than Opus 5 or GPT 5.6. This follows Fable 5's temporary shift to pay-as-you-go credits for subscribers earlier this year.

      For developers, the update represents more than just advanced AI. Enhanced coding performance, lowered operational expenses, and specialized models for high-impact sectors could position Claude as a more appealing option for developing AI-driven applications.