Anthropic has released Claude Fable 5.1, a significant update to its reasoning-focused model that delivers dramatic improvements on science benchmarks while substantially reducing inference costs.
Benchmark Performance
Fable 5.1 scores 52.6% on Terminal-Bench-Science 0.1, more than doubling its predecessor’s 24.7% score. The model also achieves 55.8% on Terminal-Bench 4.0, surpassing Mythos 5’s 42.0%. These gains represent a substantial leap in the model’s ability to handle complex scientific reasoning tasks.
A restricted variant, Mythos 5.1, shipped alongside Fable 5.1 and scores 60.9% on the coding benchmark. Access is limited to vetted US cyber-defence and life-sciences organizations, reflecting Anthropic’s cautious approach to deploying its most capable models in sensitive domains.
Cost Reductions
Anthropic has also introduced significant pricing improvements. Cache reads drop 75% to $0.25 per million tokens—roughly 25% off a typical workload and up to 45% off heavily agentic work. Input and output prices remain at $10 and $50 per million tokens respectively.
The cost reduction addresses one of the key friction points for developers building agentic applications, where repeated context lookups can quickly accumulate. By dramatically lowering cache costs, Anthropic makes long-running agent workflows more economically viable.
What This Means
The benchmark gains position Fable 5.1 as one of the strongest open-reasoning models available. The 2x improvement on science tasks suggests meaningful advances in the model’s ability to handle multi-step reasoning, domain knowledge retrieval, and complex problem-solving.
For enterprises evaluating AI models, the combination of improved reasoning and lower costs makes Fable 5.1 an attractive option for scientific analysis, research assistance, and technical documentation tasks.
The simultaneous release of Mythos 5.1 to vetted organizations indicates Anthropic continues balancing capability deployment with safety considerations—a pattern seen throughout its model release strategy.