Anthropic Releases Claude Fable 5.1 and Claude Mythos 5.1
Anthropic has launched Claude Fable 5.1 and Claude Mythos 5.1, publishing updated documentation and a joint system card. The release introduces refined writing styles, enhanced scientific reasoning capabilities, pricing updates, and technical fixes for chain-of-thought leakage vulnerabilities. This update demonstrates ongoing efforts to make LLM outputs sound less artificial while plugging critical security holes in hidden reasoning steps. Furthermore, significant price cuts in prompt caching suggest growing competitive pressures on frontier AI model API pricing. The release includes API patches to block exploits that exposed hidden chain-of-thought (CoT) traces through forced tool calls or model repetition. Additionally, prompt cache read pricing for Fable 5.1 was reduced from $1.00 to $0.25 per million tokens, cutting its cache retrieval cost to half that of Opus.
## BACKGROUND
Chain-of-thought (CoT) reasoning allows AI models to break complex problems into step-by-step logic before returning a response, though exposing these internal steps can create security and privacy risks. System cards are standard documentation released by AI research labs detailing model performance, architecture, and safety mitigations. Prompt caching is an API feature that lowers latency and cost by reusing previously computed context.