Claude Mythos: Anthropic's next-generation model leaked, 'most capable we've built'

A data leak has revealed that Anthropic is testing a new large language model called Claude Mythos, internally codenamed Capybara. Engineers have finished training it and are piloting it with early customers working on cyber defense applications.

Anthropic confirmed the model's existence, describing it as 'the most capable we've built to date.' Mythos gets dramatically higher scores on tests of software coding, academic reasoning, and cybersecurity compared to Opus 4.6.

Capybara represents a new tier of model that sits above Opus. Anthropic will add a fourth product tier when it launches Claude Mythos, which will be pricier and more capable than Opus. Think of it like: Haiku (fast/cheap), Sonnet (balanced), Opus (powerful), Capybara/Mythos (most capable).

There is no public API, no documentation, and no announced launch date yet. Only a small group of early access customers currently have access.