The strongest model for humanity, Claude Fable 5.2 and Mythos 5.2, has been released.
Almost every capability surpasses Opus 5.5.
This is currently the most cutting-edge large model for humanity, without a doubt.
In one sentence:
It is currently the strongest model for humanity, with a 20% price reduction and a 30% speed increase. Programming and office work are a breeze, and it can handle biomedicine and robotics too.
Quick notes on the key information:
1. Although it is smarter, its safety alignment is better than or on par with previous models (it may have used the "storytelling alignment" method mentioned in their research).
2. Almost all capabilities surpass Opus 5.5, and its communication style is clearer and more concise.
3. Generation speed is 30% faster than the previous generation, 5.1.
4. In the FrontierCode 2.0 test, Fable 5.2 High can tie with Opus 5.5 Max, but at half the cost per task. Note that Fable 5.2 scores highest at Xhigh, even higher than Max. Higher reasoning effort is not always better.
5. Gene therapies can be delivered via a protein shell called a capsid. Mythos 5.2 discovered a new method for predicting the entire capsid structure. It accurately modeled 28 of the 34 capsids it attempted. Biologists can use it to design capsids faster and more cheaply.
6. A team of 1,000 Fable 5.2 Agents fully formalized Fermat's Last Theorem in 30 hours, translating the proof into a language that a computer can check step by step. Under the same setup, Opus 5.5 completed only 26% in 18 hours and made almost no further progress after 30 hours.
7. Mythos 5.2 demonstrates significantly improved skill in controlling physical robots, even on tasks it was never explicitly trained for. (Like Astra, it has robotics generalization ability, which has not yet been observed in other models.) On RoboDojo, a benchmark with a simulated dual-arm robot, it outperforms other models and most AI systems built specifically for robot control.
8. The official recommendation is that for most coding and knowledge tasks, start with Opus 5.5, and hand the most ambitious, open-ended technical work to Mythos and Fable.
9. On pricing: $8 input and $40 output, down 20% from the previous $10 and $50.
These two models are different versions of the same model. Fable is positioned for the general public, while Mythos is opened up for security-sensitive areas such as biology.