🚨 Anthropic’s internal AI roadmap may have leaked

Several mysterious codenames and experimental features have reportedly appeared inside Anthropic’s development systems:

• Capybara • Fennec - (thought to be sonnet 5 but not) • Numbat - (reportedly still in testing, unknown model) • KAIROS - (possibly a persistent background agent) • ULTRAPLAN - (potentially an extended planning and reasoning mode) • Honeycomb - (widely predicted to be an Opus 5.0, but unconfirmed)

Earlier rumours linked Fennec to Sonnet 5, although newer evidence suggests that mapping may have been incorrect.

It is still unclear which names represent public models, internal checkpoints, agents or experimental Claude Code features.

Anthropic appears to have far more in development than it has officially revealed, one of these is likely Mythos 5.1/6 or fable 5.1/6 also.

What codenames do you think link to which models, and ?