Antropic has unveiled a new AI model, Claude Fable 5. Claude Fable 5 is Antropic's latest high-performance model available to general users and enterprises, focusing on complex reasoning and long-running agent tasks. Claude Mythos 5, also unveiled, shares the same performance capabilities but is available only to a limited number of customers.
The key to this model is “how reliably it can handle time-consuming and complex tasks.” Antropic explained that Claude Fable 5 can work autonomously for longer periods than the previous Claude model, and that performance has improved in software engineering, knowledge work, vision, memory, and life science research.
Changes are particularly significant in the development domain. Antropic introduced Claude Fable 5 as suitable for large-scale code migrations, complex implementations, and long-term projects spanning several days. GitHub also announced the introduction of Claude Fable 5 to GitHub Copilot, rolling it out sequentially to Copilot Pro+, Max, Business, and Enterprise users. It can be used as a model selector across various environments, including VS Code, Visual Studio, JetBrains, Xcode, and GitHub Mobile.
As noteworthy as its performance is the safeguards. Claude Fable 5 includes a secure classifier capable of rejecting certain sensitive requests. In contrast, this classifier is not applied to Claude Mythos 5, the restricted-access model. Antropic explained that it applied more conservative protections to the general-access model, recognizing the risk that high-performance models could be exploited in sensitive fields such as cybersecurity, biology, and chemistry.
This safeguard also sparked controversy because the process of some requests being rejected or rerouted to other models was not sufficiently transparent to users. Subsequently, Antropic admitted that it had “chosen the wrong balance” and stated that it would notify users more clearly when Claude Fable 5 rejects or reroutes requests.
Based on API, Claude Fable 5 and Mythos 5 support a default 1 million token context window and a maximum output of 128,000 tokens. Pricing is proposed at $10 per 1 million token input and $50 per 1 million token output. Both models will be available starting June 9, 2026, and Claude Fable 5 will be generally available on the Claude API, AWS, Amazon Bedrock, Vertex AI, and Microsoft Foundry.
More and Sources
