Anthropic has released Claude Fable 5.1 and Claude Mythos 5.1

The base cost of generation remains unchanged ($10/$50 per million tokens), but cache reads have become 75% cheaper ($0.25 per 1 million tokens).

The model performs better in long-term agent tasks in coding, multi-step research, as well as working with documents, spreadsheets, and presentations.

According to Anthropic’s recommendation, Claude Opus 5 is sufficient for most tasks, while Fable 5.1 should be used when deeper reasoning is required or when Opus fails even at maximum settings.

New features include the ability to adjust the “effort” level for specific messages, temporary system prompts within a single turn, and readable progress statuses between tool calls—all three are currently in beta.

Mythos 5.1 has the same features and pricing but is available only to participants of the Project Glasswing program by invitation.