Anthropic Launches Fable 5.1, Boosts Research and Coding, Cuts Costs by 45%

On September 1, Anthropic officially launched Claude Fable 5.1 and Claude Mythos 5.1, marking a decisive shift in the generative AI landscape from benchmark optimization to real-world problem solving. The release underscores a broader industry transition where competitive advantage now hinges on computational efficiency, specialized safety controls, and demonstrable scientific utility. Fable 5.1 delivers substantial performance upgrades, doubling its score on the Terminal-Bench-Science 0.1 benchmark to 52.6% while improving coding metrics by 13%. Beyond standard tasks, the model has demonstrated advanced capabilities in computational biology by designing high-affinity protein binders with a 50% success rate, significantly outperforming previous results. The system also successfully rewrote GPU kernel code for seven open-source deep learning frameworks, accelerating inference speeds by up to 2.5x and reducing compute expenses for research labs by as much as 60%. In planetary science, Fable 5.1 processed decades-old NASA radar imagery to generate a high-resolution topographical map of Venus, advancing spatial detail from the previous 10-to-20-kilometer range down to 2-to-3-kilometer precision. Commercially, Anthropic has substantially restructured its pricing to attract enterprise and developer adoption. Fable 5.1 reduces cache read costs by approximately 75%, bringing the price to $0.25 per million tokens. Standard operational costs have decreased by 25%, with complex multi-agent workflows seeing savings of up to 45%. This aggressive pricing strategy is underpinned by massive infrastructure commitments, including a $35 billion computing lease with Lambda and a $45 billion agreement with Nscale. The expansion includes a new data center deployment in Texas, structured to mitigate financial risk through partnerships with Hut 8 Corp. and NVIDIA. The release introduces a dual-variant architecture to address varying compliance requirements. Fable 5.1 is available to all users, while Mythos 5.1 is restricted to vetted institutions and features relaxed safety thresholds tailored for sensitive cybersecurity and life sciences research. Anthropic reports that safety interventions for biological and defensive cybersecurity tasks have dropped by 85% and 60% respectively. For enterprise clients, the company is rolling out Enterprise Frontier Safeguards, a zero-data-retention solution that processes information on customer-hosted cloud environments while maintaining model-level abuse detection. Concurrently, Anthropic has implemented strict anti-distillation protocols that prevent new API accounts from manipulating conversation history to extract internal reasoning processes. The simultaneous deployment of advanced scientific tools, aggressive cost reductions, and tiered security frameworks signals a maturation in AI development. As model capabilities plateau across standard benchmarks, the industry is now competing on infrastructure scale, economic efficiency, and domain-specific reliability, establishing Fable 5.1 as a benchmark for practical, enterprise-grade artificial intelligence.