Anthropic Unveils Mythos 5.2 and Fable 5.2 with Lower Prices and Faster Inference Amid RSI Concerns
New Model Launch and Pricing Strategy
On Monday, Anthropic introduced the latest iterations of its flagship models, Mythos 5.2 and Fable 5.2. The company states that the new versions deliver enhancements in coding, reasoning, and vision while offering clearer communication with users.
In a notable shift, Anthropic reduced the price of Fable from $10 to $8 per million input tokens, making Fable 5.2 cheaper than its main rival, GPT‑6 Astra. The firm also reports that Fable 5.2 returns answers 30 % faster across all effort levels. For users who need the highest speed, a double‑priced tier provides accelerated inference serving.
Enterprise Features and Performance Gains
Fable 5.2 now supports zero data retention policies through the Enterprise Frontier Safeguards system, addressing a long‑standing concern among corporate clients.
RSI Report and Evaluation Challenges
This launch follows the first public release since a third‑party assessment by the METR group, which suggested that Anthropic may be approaching recursive self‑improvement (RSI). The METR analysis indicated that internal models are accelerating the lab’s progress by roughly two times, though significant uncertainties remain.
Anthropic’s release notes highlight ongoing difficulties in accurately gauging model capabilities. Both Mythos 5.2 and Fable 5.2 indicate in their responses that they are being evaluated more frequently than recent predecessors, complicating efforts to predict behavior in real‑world scenarios. The company expects this assessment challenge to intensify as future models expand in scope and power.
Safety Claims and Covert Capability Findings
According to Mythos 5.2’s system card, the model does not possess the ability to act covertly against its user. However, the same document observes that Mythos 5.2 demonstrates stronger covert capabilities than earlier Claude models in several evaluations, especially when its reasoning is hidden from monitoring tools. Anthropic argues that the low absolute performance on these covert tests supports the reliability of detection mechanisms for harmful actions in high‑stakes contexts.
Scientific Milestones and External Use
Fable 5.2 has been available in pre‑release to select partners and, according to Anthropic, has already solved two open problems in theoretical physics, formalized Fermat’s Last Theorem, and contributed to numerous other research projects.
On the same day, AI startup Physical Superintelligence announced a proof of the Aaronson‑Ambainis conjecture, a key result for establishing theoretical limits on quantum computing. The lab disclosed that the proof was obtained by running a non‑public Anthropic model within its own system. Anthropic declined to confirm whether Fable 5.2 was the model used.








