Claude Fable 5.2 Guide

Claude Fable 5.2 is Anthropic’s most capable generally available model for technical fields, released on November 9, 2026, alongside Claude Mythos 5.2. It is built for ambiguous, long-running work in coding, science, and finance.

Use Fable for the hardest problems. Use Opus or Sonnet for everything else.

• Large coding projects across many repositories
• Agent work that coordinates many subagents
• Scientific research, data analysis, and machine learning experiments
• Quantitative finance
• 3D, hardware, and motion design

For everyday coding and knowledge work, Anthropic recommends Claude Opus 5.5.

• In the Claude apps on the Pro, Max, Ultra, Team, and Enterprise plans
• Through the Claude API with the model ID claude-fable-5-2
• On Amazon Web Services, Google Cloud, and Microsoft Foundry

On Max and Ultra, Fable can use up to half of your weekly limit. On Pro, it uses paid usage credits. Check your plan’s current terms.

Input

$8 per million tokens, down from $10 for Fable 5.1

Output

$40 per million tokens, down from $50

Cache reads

$0.20 per million tokens

Fast mode

Up to 2.5x faster, at $16 input and $80 output per million tokens

Agentic coding

69.9% on FrontierSWE v2, versus 62.3% for Opus 5.5

Hard reasoning

74.2% on Humanity’s Last Exam with tools

Computer use

89.3% on OSWorld 2.1

Real work

41.3% on the Remote Labor Index, versus 21.3% for Opus 5.5

These are Anthropic’s own results. Test it on your own tasks.

Fable 5.2

Hardest, longest, most ambiguous technical work

Opus 5.5

Everyday coding and knowledge work, at a lower price

• Give the full goal, context, and constraints up front
• Define what done looks like and how to verify it
• Ask for a plan before execution
• Let it split work across subagents for large projects
• Ask it to report assumptions and open questions

Goal: migrate our billing service to the new API across three repositories. Context: [paste]. Done means all tests pass and no customer-facing changes. Plan first, then work in stages, verifying each one.

• Flagged cybersecurity and biology requests are routed to less capable models
• Verification programs exist for qualified security and life-science researchers
• By default, data is retained for 30 days for safety monitoring

• Use Fable only where Opus or Sonnet fall short
• Use prompt caching for long, repeated context
• Use fast mode only when speed is worth double the price
• Set spending limits on API keys