Claude Fable 5.2

Claude Fable 5.2 is Anthropic's current Fable-class model for difficult coding, finance, and science work. Released on November 9, 2026, its API ID is claude-fable-5-2. Anthropic positions it for long-horizon agent sessions, multi-step research, and work where a smaller or daily-driver model has already failed an evaluation.

Specifications

Anthropic says fast mode can deliver up to 2.5x faster output at $16/$80 per MTok. It also documents four migration-affecting differences from 5.1, including incompatible thinking blocks and stricter prefix-check behavior. Test the new model ID before changing production defaults.

Practical workflow

Start with a representative repository or research task and measure completion quality, elapsed time, token cost, and tool failure recovery. Use the effort control to limit unnecessary depth, keep a cheaper fallback for routine jobs, and review any safety or permission boundary outside the model. Anthropic recommends starting with the latest Opus model for most workloads and moving to Fable when higher-effort evaluation results remain insufficient.

Fable 5.2 is not automatically a cheaper replacement for every task: its list price is lower than Fable 5.1 but materially above lightweight models, and adaptive thinking cannot be disabled. For the prior model see Claude Fable 5.1; for a daily generalist compare Claude Opus 5.5.

Conclusion

Use Fable 5.2 for tasks whose ambiguity, duration, or reasoning depth warrants its cost. Confirm the current limits and migration notes in Anthropic's official model documentation before deployment.