Anthropic Releases Claude Fable 5.1 and Mythos 5.1 with 75% Lower Cache Fees
Anthropic officially announced Claude Fable 5.1 and Mythos 5.1, featuring notable performance gains across coding, knowledge work, and scientific research benchmarks. Alongside the launch, Anthropic reduced prompt cache read fees by 75% down to $0.25 per million tokens. The price reduction in prompt caching lowers overall operating costs by up to 45% for highly agentic workflows, making complex AI deployments far more affordable. Additionally, the release demonstrates progress in domain-specific AI for biology and cybersecurity while aligning with regulatory standards like the EU AI Act. Fable 5.1 scored 55.8% on the Terminal-Bench 4.0 coding benchmark and 52.6% on Terminal-Bench-Science 0.1, surpassing GPT-5.6 Sol and Opus 5. Anthropic also introduced Enterprise Frontier Safeguards (EFS) for cloud-level zero data retention and implemented invisible text watermarking to comply with EU regulations.
## BACKGROUND
Prompt caching in large language models allows systems to reuse pre-processed context prefixes, such as system prompts or large documentation, reducing both latency and API costs. Benchmarks like Terminal-Bench measure an AI agent's ability to execute complex technical tasks inside real terminal environments.