Anthropic has launched Claude Fable 5.1, significantly outperforming its predecessor and introducing cost reductions for users.
Anthropic Launches Advanced AI Models
On September 1, Anthropic introduced two new AI models, Claude Fable 5.1 and Claude Mythos 5.1, which have shown remarkable improvements over their predecessors. The Fable 5.1 achieved a score of 52.6% on the Terminal-Bench-Science 0.1 benchmark, a significant increase from Fable 5's score of 24.7%. Similarly, on the Terminal-Bench 4.0 test, it scored 55.8%, up from 42.0% in the previous model.
Cost Reductions and Access
Alongside performance enhancements, Anthropic has reduced the cost of cache reads by 75%, leading to an overall workload cost decrease of approximately 25%. For more intensive workloads, this reduction can be as high as 45%. Despite these changes, the pricing structure remains the same at $10 per million tokens for input and $50 for output.
Fable 5.1 is available to all Claude account holders, while Mythos 5.1 is restricted to professionals in cybersecurity and life sciences who are part of Anthropic's verification programs.
Performance Metrics and Comparisons
Performance metrics are crucial for evaluating these AI models. Fable 5.1 not only outperformed its predecessor but also surpassed the recently launched Opus 5 in all significant benchmarks. Opus 5 had initially posed a competitive challenge due to its lower pricing and better performance on some tests. However, Fable 5.1's enhancements have restored its leading position.
In terms of academic applications, Fable 5.1 scored 60.9% on the Humanity's Last Exam, confirming its capabilities in multidisciplinary reasoning.
Access and Availability
The structure for accessing Fable 5.1 remains similar to that of Fable 5, with certain plans including it as a standard offering, while others require pay-as-you-go credits. The new model is now accessible through the Claude API and various cloud platforms, including Amazon Bedrock and Google Cloud.
