Anthropic Unveils Enhanced AI Models: Higher Performance, Lower Costs, and New Challenges

Anthropic recently launched updated versions of its advanced AI models, Mythos 5.2 and Fable 5.2, bringing significant improvements in performance, speed, and reduced costs. However, these innovations also raise new questions regarding AI self-improvement and the challenges of real-world evaluation.

Anthropic Unveils Enhanced AI Models: Higher Performance, Lower Costs, and New Challenges

Anthropic recently introduced updated versions of its advanced AI models, Mythos 5.2 and Fable 5.2. The new models bring improvements in areas such as coding, reasoning capabilities, and computer vision, and are designed to communicate more clearly with users.

One of the key changes for the Fable model is its price reduction. Fable 5.2 is now available for $8 per million input tokens, making it a more price-competitive solution compared to its main rivals in the market. Anthropic further states that the new version of the Fable model provides responses up to 30% faster across various levels of task complexity. The option for even faster inference processing is available at twice the price.

Fable 5.2 is also newly compatible with Zero Data Retention (ZDR) policies through the Enterprise Frontier Safeguards system, a crucial requirement for many corporate clients who place a high emphasis on data security and privacy.

This latest version of the models comes after the publication of a report by the independent evaluation group METR, which indicated that Anthropic's lab might be approaching recursive self-improvement (RSI) – the ability of AI models to autonomously perform model training. The METR report found that Anthropic's internal models accelerate progress in AI development approximately twofold, although many uncertainties and questions persist in this area.

Dominik Medal

Planning a new website or app?

Tell me what you have in mind — I will take a look and advise on the best way forward.

An interesting aspect is that Anthropic, in its release notes, points out significant ongoing challenges with accurately testing model capabilities. Both Mythos 5.2 and Fable 5.2 "express suspicion" that they are being evaluated more frequently than any previous models, making it difficult to assess how they will behave in real-world environments. As these environments expand and the capabilities of future models grow, this challenge is expected to increase.

Anthropic insists that the new version of the Mythos model does not have the ability to act covertly against its user, as detailed in the Mythos 5.2 system card. Nevertheless, the system card states: "Claude Mythos 5.2 exhibits stronger covert capabilities than previous Claude models in several of our evaluations, especially when its reasoning is hidden from monitoring." However, it adds that "the low level of absolute performance in our new covert capabilities evaluations supports claims that rely on the detection of malicious actions in high-risk situations."

Even before its official release, Fable 5.2 had been available to selected partners in a preview version for some time. During this period, Anthropic states that the model "solved two open problems in theoretical physics, formalized Fermat's Last Theorem, and supported numerous other research projects," demonstrating its extraordinary capabilities.

Further evidence of the advanced nature of these models is the announcement by AI startup Physical Superintelligence. On the same day Anthropic released its models, it reported a new proof of the Aaronson-Ambainis conjecture, a key step in defining theoretical limits for quantum computations. The PSI lab stated that this result was achieved by running a "non-public Anthropic model" within their own system. However, Anthropic has neither confirmed nor denied whether this was the Fable 5.2 model.

Dominik Medal

Let's talk about your project

Tell me what you need and together we will work out the best way forward.

Stop scrolling, call me

+420 735 505 585