Anthropic announced the launch of Claude Fable 5.1 and Claude Mythos 5.1, its latest large language models designed for advanced coding and knowledge work. The company claims these models deliver stronger performance than their predecessors and introduce significant cost efficiencies, particularly for complex agentic workloads. Anthropic states that Claude Fable 5.1 reduces the cost of typical workloads by approximately 25% and highly agentic workloads by up to 45% compared to Fable 5. This reduction primarily stems from a 75% decrease in pricing for cache reads, which are now $0.25 per million tokens, down from $1.00 for Fable 5.
Claude Fable 5.1 is generally available to Pro, Max, Team, and Enterprise users, as well as developers building AI solutions through the Claude Platform, Amazon Web Services, Google Cloud, and Microsoft Foundry. Claude Mythos 5.1 is the same underlying model but features different safeguard levels and is available through restricted-access programs for vetted cybersecurity and life-sciences organizations. These specialized programs allow access to capabilities that are typically constrained by the production safeguards present in Fable 5.1.
The base API rates for Fable 5.1 remain consistent with Fable 5, priced at $10 per million input tokens and $50 per million output tokens. While these rates are higher than other models in Anthropic's lineup, such as Opus 5 ($5 per million input, $25 per million output) and Sonnet 5 ($2 per million input, $10 per million output), the reduced cache read pricing aims to make sustained agentic operations more economical. For Fable 5.1 and Mythos 5.1, a cache hit now costs 2.5% of the standard input price, a decrease from the 10% multiplier used by most other Claude models.
Anthropic has also addressed customer feedback regarding data retention and model safeguards. The company is introducing Enterprise Frontier Safeguards (EFS), a system that allows eligible enterprise customers to store data within their own cloud infrastructure, effectively providing zero data retention from Anthropic's perspective. Until EFS becomes broadly available later this fall, eligible customers can use Fable 5.1 with zero data retention. Previously, Fable models typically required 30-day data retention for safety monitoring, which had been a point of concern for some users.
Improvements have also been made to the models' safeguards. Anthropic reports that Fable 5.1's cybersecurity safeguards produce approximately 60% fewer false positives than before. This enhancement permits Fable 5.1 to be used for discovering software vulnerabilities, though not for developing exploits. For biology-related tasks, a new access program has been established, in partnership with the US government, to enable access to Mythos 5.1's advanced biology capabilities, with open enrollment for scientists expected soon.
Anthropic positions Fable 5.1 for sustained problem-solving and complex, multi-stage knowledge work. The model is designed for tasks that may span hours and involve multiple applications, such as working through backlogs, responding to requests in communication platforms, operating a browser, or running as an unattended agent. For coding projects, Fable 5.1 is described as Anthropic's most capable model, able to handle features across an entire codebase, perform code reviews, and conduct multi-day autonomous sessions. It can also write its own tests and use vision capabilities to verify outputs against design goals.
The company highlights Fable 5.1's ability to understand diagrams, charts, and tables embedded in files and PDFs, which is intended to improve document-heavy work in sectors like finance, legal, analytics, and architecture. Anthropic states that Fable 5.1 exhibits improved judgment on ambiguous tasks and provides fewer confident incorrect answers compared to Fable 5. The model also includes invisible watermarking for generated text, with a detection API being rolled out in a private preview for eligible organizations.