Anthropic announced two new Claude models today for challenging software development, research and knowledge-intensive tasks. Claude Fable 5.1 is now available for the general public, and Claude Mythos 5.1 is being provided to select customers via Anthropic’s verification programs. Both models are based on the same underlying system, but Anthropic has applied different safety controls to each version.
Mythos 5.1 is intended for approved organisations operating in domains such as cyber security and life sciences where researchers may need to access capabilities that are limited in the standard version.
Claude Fable 5.1 and Mythos 5.1: What's New
The two models were detailed in an Anthropic blog post as being models that are suited for tasks that involve programming, knowledge-related work and reasoning. The creation of the models was informed, in part, by user feedback on pricing, data holding and safety restrictions. Claude Mythos 5.1 is similar to Fable 5.1 in its core abilities and has a less restricted safety configuration that is available to those that have been approved by Anthropic. Access to this model will be via the Cyber Verification Program and Life Sciences Verification Program.
Anthropic is also touting the cost effectiveness of the new Fable 5.1 version. The cost, according to the company, is around 25 percent cheaper than Fable 5 due to reduced cache reads. In cases where there is heavy agentic use, it can even be as low as 45 percent cheaper.
Claude Fable 5.1 Benchmark Results
Based on the internal testing by Anthropic, Fable 5.1 has made major strides over its predecessors when it comes to scientific reasoning and agentic coding. For instance, in Terminal-Bench-Science test, Fable 5.1 scored 52.6 percent compared to Fable 5 which got 24.7 percent while Opus 5 scored 29 percent.
We’re introducing Claude Fable 5.1 and Claude Mythos 5.1.
— Claude (@claudeai) They're the world’s most advanced models for coding and knowledge work.[pic.twitter.com/8P9PSrWPi3][September 1, 2026]
In CursorBench test, the model also scored 73.4 percent compared to Fable 5 that scored 70.5 percent and Opus 5 which scored 70 percent, making Fable 5.1 better than its predecessors in CursorBench test. Fable 5.1 has also gone through real-world testing in addition to benchmark tests according to Anthropic. In an experiment carried out using the help of investment firm Millennium, Fable 5.1 successfully identified the problem causing persistent software crash to be a bug in a third party library.
The organization has put the model through more testing in fields like computational analysis and modeling, scientific research, and computational biology. According to Anthropic, Mythos 5.1 has had a success rate near 50 percent in the design of proteins with high affinity.
Also See: NYC Mayor Mamdani Bans AI For Students Up to Class 8: Why Is He Limiting Gen AI in Schools