ReleaseBytes
anthropic Claude Platform Release Notes ·

Anthropic Launches Claude Fable 5.1 and Mythos 5.1 with Advanced AI Capabilities

aipreviewengineer
feature

Anthropic has launched Claude Fable 5.1 and Claude Mythos 5.1, offering advanced capabilities for agentic coding and knowledge work. Both models feature a 1M token context window, 128k max output tokens, and always-on adaptive thinking, with pricing consistent with Claude Fable 5 but reduced prompt cache read costs. Claude Fable 5.1 is generally available across multiple platforms, while Mythos 5.1 is for Project Glasswing participants. The release also introduces several beta features, including per-message effort configuration and turn-scoped system messages.

  • →Launch of Claude Fable 5.1 and Claude Mythos 5.1
  • →Strict Tool Choice Enforcement and Thinking Block Logic
  • →New Beta Features for Conversation and Agent Control
  • →Optimized Pricing with Reduced Prompt Cache Costs
  • →Enhanced Content Integrity and Data Handling
Breaking changes (1) ›
  • Strict Tool Choice Enforcement and Thinking Block Logic

    The `tool_choice` types `any` and `tool` are no longer supported on Claude Fable 5.1 and Mythos 5.1, requiring strict tool use or structured outputs. Thinking blocks are now model-version-dependent, preserved only for the producing model or newer, with new API checks for history consistency.

Features (2) ›
  • Launch of Claude Fable 5.1 and Claude Mythos 5.1

    Anthropic has launched Claude Fable 5.1 and Claude Mythos 5.1, successors to Claude Fable 5, designed for long-running agentic coding, knowledge work, and research. Both models support a 1M token context window, 128k max output tokens, and always-on adaptive thinking. Claude Fable 5.1 is generally available, while Mythos 5.1 is for Project Glasswing participants.

  • New Beta Features for Conversation and Agent Control

    Several new beta features are available, including per-message effort changes to modify model behavior mid-conversation, and turn-scoped system messages that provide temporary context without accumulating token cost. Additionally, `thinking.display` now accepts an "updates" value to show short progress updates during agentic tool calls.

Enhancements (1) ›
  • Optimized Pricing with Reduced Prompt Cache Costs

    Pricing for Claude Fable 5.1 and Mythos 5.1 remains consistent with Claude Fable 5 at $10/$50 USD per MTok for input/output. However, prompt cache reads now cost $0.25 USD per million tokens, representing a 0.025x reduction compared to the base input price.

Notes (1) ›
  • Enhanced Content Integrity and Data Handling

    Text generated by Claude Fable 5.1 and Mythos 5.1 now includes Anthropic's text watermark, and media produced via the code execution tool carries C2PA Content Credentials. Both models continue to require 30-day data retention, with zero data retention only available by special authorization.

Read the original announcement →

https://platform.claude.com/docs/en/release-notes/overview

Related releases