Anthropic released Claude Fable 5.1 and Claude Mythos 5.1, which are the same underlying model with different safety guardrails. Fable 5.1 is publicly available; Mythos 5.1 has stricter safeguards for cybersecurity and life sciences work and requires trusted access. The headline improvements are price cuts (25% cheaper for typical workloads, up to 45% for agentic tasks), new data privacy options through Enterprise Frontier Safeguards that store data on customer infrastructure instead of Anthropic's servers, and reduced false positives in safety filters (60% fewer in cybersecurity). The models outperform their predecessors across benchmarks—Fable 5.1 scores 55.8% on agentic coding versus Fable 5's 42%, and 52.6% on scientific research versus 24.7%. The real story is the scientific capabilities. Mythos 5.1 designed protein binders with affinities 10 times higher than competition entries, with a 50% hit rate (versus the typical 10-15%). It created a high-resolution elevation map of Venus from 30-year-old radar data, revealing details at 2-3 kilometers instead of 10-20. It also optimized deep learning models by writing custom GPU code, speeding up seven open-source models by up to 2.5x and cutting GPU costs by 30-60% on large-scale experiments—work that would normally take performance engineers weeks and is often unaffordable for academic labs. Anthropic is betting these models will contribute meaningfully to scientific discovery. They're offering free credits through an AI for Science program, discounted access via Claude Team for researchers, and recently previewed hardware that lets Claude operate laboratory equipment directly. They've also invested in safety work—the models underwent extensive testing for chemical and biological risks before release, with a full System Card documenting their approach. The company argues that as AI autonomy increases, safety and alignment efforts need to keep pace.