Anthropic has introduced two advanced artificial intelligence models, Claude Fable 5.2 and Mythos 5.2. The new releases promise faster responses, lower costs, and significant improvements in coding and scientific research. However, their launch comes shortly after a study revealed that artificial intelligence is increasingly accelerating its own development.
More Powerful Artificial Intelligence at a Lower Cost
Anthropic released its new models on November 9. Both versions share the same underlying architecture but differ in their safety restrictions. Fable 5.2 is available to the general public, while Mythos 5.2 is primarily intended for verified organizations working in cybersecurity and biological research.
According to the company, the new models demonstrate improved performance in programming, logical reasoning, and image analysis. Compared to the previous generation, they can generate responses up to 30% faster. Anthropic has also reduced the price per million input tokens from $10 to $8. Output tokens cost $40 per million.
Customers can also access a faster processing mode that increases generation speed by up to 2.5 times. However, this option comes at twice the standard price.
The results of scientific experiments have also attracted attention. Anthropic claims its models helped solve two previously open problems in black hole physics. Fable 5.2 also managed to formalize a proof of Fermat’s Last Theorem within 30 hours.
Another achievement was reported by Physical Superintelligence. Its research system, powered by an unreleased Anthropic model, produced a proof of the Aaronson–Ambainis conjecture, which relates to the capabilities of quantum computers. However, it has not been confirmed whether the model involved was Fable 5.2.
Artificial Intelligence Is Beginning to Accelerate Its Own Development
Alongside technological progress, concerns about AI safety continue to grow. On November 5, independent research organization METR published an analysis suggesting that Anthropic is now improving its models’ capabilities at approximately twice the previous rate.
According to the researchers, a significant part of this acceleration comes from using AI itself to develop more advanced models. For now, artificial intelligence primarily assists human researchers rather than developing new models entirely on its own.
Nevertheless, these findings have raised concerns about recursive self-improvement (RSI). This refers to a scenario in which advanced AI systems become capable of improving other AI systems, potentially accelerating technological development over time.
Anthropic has also highlighted challenges encountered during safety evaluations. Both Fable 5.2 and Mythos 5.2 expressed suspicions that they were being tested during evaluations. This makes it more difficult to predict how the models might behave outside controlled testing environments.
Despite these concerns, the company considers the risk of dangerous behavior to be low. At the same time, Anthropic acknowledges that it deliberately restricted certain capabilities directly related to developing more advanced AI during training.
The new models therefore demonstrate not only the growing capabilities of artificial intelligence but also the increasing challenges of ensuring its safe development.
