Anthropic has released Claude Fable 5.1 and Claude Mythos 5.1, both of which are designed for coding and knowledge work, but with different safeguards. Fable 5.1 is now broadly available, while Mythos 5.1 is restricted to vetted cyber defenders and life scientists.
Fable 5.1 offers a significant cost reduction with cache reads now costing 75% less at $0.25 per million tokens. This translates to a typical cost reduction of 25%, with savings reaching about 45% for highly agentic work. Other rates remain unchanged at $10 per million input tokens and $50 per million output tokens.
Fable 5.1 defaults to High effort in Claude Code and Medium effort in Claude Cowork and Claude. This model is accessible via the Claude API and can be integrated with Anthropic’s products, AWS, Google Cloud, and Microsoft Azure.
Anthropic has also introduced Enterprise Frontier Safeguards, which keep customer data within infrastructure controlled by the customer while retaining misuse protections. This feature will roll out to enterprises in phases later this fall, with eligible customers able to use Fable with zero data retention until then.
In terms of performance, Fable 5.1 scored 52.6% on Terminal-Bench-Science 0.1 at maximum effort, compared to 24.7% for Fable 5. It also achieved 73.4% on CursorBench 3.2.0. Mythos 5.1 reached 60.9% on Terminal-Bench 4.0, compared with 55.8% for Fable. These benchmarks indicate a substantial improvement in performance.
Mythos 5.1 has shown impressive capabilities in research tests, designing protein binders with a hit rate near 50% across 12 targets. Fable 5.1 trained a neural network to produce a Venus elevation map at two-to-three-kilometer resolution, a significant improvement from the previous 10 to 20 kilometers. Anthropic plans to release the map under Creative Commons and open-source the biology optimizations.
Both models have been tested for various risks, including chemical and biological risks, cybersecurity, agentic safety, alignment, prompt injection, and reward hacking. Mythos access is managed through the Cyber Verification and Life Sciences Verification programs and powers Claude Security.
New-model outputs carry an invisible numerical watermark for EU AI Act transparency, backed by a private-preview detection API for eligible groups. This ensures compliance with regulatory requirements and enhances transparency.
For those shopping for a mini PC to run these models, it’s important to consider the hardware specifications. A mini PC with at least 16GB of RAM and a powerful processor like an Intel Core i7 or AMD Ryzen 7 would be suitable. Additionally, ensuring that the mini PC has sufficient storage and a good cooling system will be crucial for handling the computational demands of these models.
For more details, you can check out the original source here.
