Claude Fable 5 and Claude Mythos 5 launched with 1M token context, adaptive thinking, and new refusal categories

Claude Fable 5 (claude-fable-5) and Claude Mythos 5 (claude-mythos-5) launched with 1M token context window by default, 128k max output tokens, and always-on adaptive thinking. Both use the new tokenizer from Claude Opus 4.7 (producing ~30% more tokens). Claude Fable 5 runs safety classifiers and returns stop_reason: "refusal" on declined requests; opt-in fallbacks parameter re-runs refused requests on another model. New stop_details.category field includes "reasoning_extraction" for requests blocked under reverse engineering restrictions. On both models, adaptive thinking is the only thinking mode; thinking.display defaults to "omitted". Claude Fable 5 requires 30-day data retention and is not available under zero data retention.