ANTHROPIC released Fable 5.1, saying the model improves performance on programming and scientific tasks while lowering enterprise operating costs. The model is positioned for long, complex software tasks — including full-application development and code review — and expands scientific capabilities such as experimental design, result simulation and reading complex charts and tables. Enterprise token pricing remains unchanged from Fable 5, but Anthropic cut pricing for re-invoking already-processed context by 75% — a usage segment that can exceed half of tokens in long-text or complex workflows — which Anthropic says should materially lower overall inference costs. Fable 5.1 also upgrades its safety classifier to better detect risky instructions and reduce false positives, targeting safer use in cybersecurity and bio-related applications. Anthropic plans to allow enterprise customers to store strongest-model data on their own cloud infrastructure rather than Anthropic servers, while continuing to apply safety checks.(金十数据)