Anthropic launched on Tuesday (1st) the models Claude Fable 5.1 and Claude Mythos 5.1, with performance improvements and changes aimed at two recurring customer complaints: high cost and safeguards that blocked legitimate requests. The company estimates that Fable 5.1 can cost up to 45% less on highly agentic workloads.
The main reduction came from cache reads, a mechanism used when the model reuses information it has previously processed. The price dropped 75%, to US$ 0.25 per million tokens. According to Anthropic, this reduces the cost of typical workloads by approximately 25% and can reach about 45% on complex programming tasks and intensive agent use.
The other prices remain the same as those of Fable 5: US$ 10 per million input tokens and US$ 50 per million output tokens.
The change is already altering model choices at some companies. Cognition, responsible for the Devin programming agent, said it intends to transfer part of the traffic that currently uses Opus 5 to Fable 5.1.
Walden Yan, cofounder and CPO of the company, said that “a Fable-class model has finally become economically viable for workloads we kept on Opus”. In the company's internal tests, the new model matched or surpassed Fable 5 with lower cost per task.
Fewer blocks also affect performance
Another relevant change is in safeguards. Anthropic says the new cybersecurity protections cause about 60% fewer interventions per session in Claude Code than those used in Fable 5.
Fable 5.1 can also now be used to identify vulnerabilities in software, although tasks considered dual-use, such as exploit generation and certain penetration tests, continue to be directed to models with other access policies.
This protection system has a measurable impact on performance itself. Fable 5.1 and Mythos 5.1 use the same base model, but have different levels of safeguards. On Terminal-Bench 4.0, focused on programming tasks executed by agents, Fable 5.1 scored 55.8%, while Mythos 5.1 reached 60.9%.
Anthropic itself attributes the difference to tasks in which Fable's more restrictive safeguards intervened and says it expects that gap to narrow with the new protections.
Mythos 5.1, however, will not have broad distribution. The version is aimed at previously vetted professionals and organizations working in sensitive areas, mainly cybersecurity and life sciences. Meanwhile, Fable 5.1 is generally available through the Anthropic API and on platforms such as AWS, Google Cloud, and Microsoft Azure.
The company is also preparing Enterprise Frontier Safeguards for corporate clients, a system in which data remains on infrastructure controlled by the client itself. Deployment will begin in phases later this Northern Hemisphere autumn; eligible clients will be able to use zero data retention until the system becomes available.
