Anthropic has released two updated models, Claude Fable 5.1 and Claude Mythos 5.1, bringing measurable gains in scientific reasoning and a significant reduction in caching costs for API users. The release adds incremental but concrete improvements to the existing Claude 5 generation, with the headline figures being a 52.6% score on Terminal-Bench-Science and a 75% drop in cache read pricing.
What Terminal-Bench-Science Measures
Terminal-Bench-Science is a benchmark designed to test a model's ability to carry out multi-step scientific tasks in a terminal environment, covering areas such as data analysis, experimental reasoning, and code-driven problem solving. A score of 52.6% represents a meaningful step forward for the Claude 5 line, which began with the original Fable 5 release earlier this year. The benchmark is considered demanding because it requires sustained, accurate reasoning across long task chains rather than isolated question answering.
Key Facts
- Claude Fable 5.1 scores 52.6% on Terminal-Bench-Science
- Cache read prices reduced by 75% for both models
- Both models available via the Anthropic API
- Updates are incremental releases within the Claude 5 generation
- Improvements target scientific reasoning and developer cost efficiency
Fable 5.1 is positioned as the more capable of the two updated models, while Mythos 5.1 continues to serve use cases where lower latency and cost are priorities. When Fable 5 and Mythos 5 launched together, the two-model approach was already framed around that capability-versus-efficiency trade-off, and the 5.1 updates preserve that structure.
The 75% reduction in cache read costs is one of the more direct signals that Anthropic is responding to developer feedback about running large-context workloads at scale.MarkTechPost
The Cache Pricing Change and Its Practical Impact
For developers building applications that rely on prompt caching, the cost reduction is the more immediately actionable update. Prompt caching allows repeated context, such as long system prompts or large documents, to be stored and reused rather than reprocessed on every call. Cutting the cache read price by 75% lowers the per-token cost of those reused inputs substantially, which matters most for applications making frequent API calls with shared context. A detailed breakdown of the cache cost changes outlines exactly how the new pricing tiers apply to different usage patterns.
The update arrives alongside ongoing interest in Claude's performance on rigorous scientific and agentic tasks. Anthropic's earlier debut of Claude Science signaled the company's intent to compete on technical and research-oriented benchmarks, and the Fable 5.1 Terminal-Bench-Science result fits that broader direction. Whether the 52.6% score translates into real-world gains for researchers using the API will depend on the specific tasks involved, but the number does move the needle compared to prior Claude 5 checkpoints.
Availability and What Comes Next
Both Claude Fable 5.1 and Claude Mythos 5.1 are available now through the Anthropic API. Developers already using Fable 5 or Mythos 5 can access the updated versions through the standard model selection process. Anthropic has not announced a separate pricing change for input or output tokens beyond the cache read reduction, so the primary financial benefit is concentrated in high-caching use cases.
The .1 versioning suggests these are refinements rather than architectural overhauls, consistent with how the company has iterated on previous model generations. For teams tracking Claude's model family across capability tiers, Fable 5.1 and Mythos 5.1 occupy the same positions as their predecessors, now with better science benchmark performance and lower operational costs for cached workloads.
