Skip to content
ForkLog·Indexed by this site · Originally published· AI updates →

Anthropic limits Claude Fable 5.2's ability to accelerate the development of new AI models

Anthropic limits Claude Fable 5.2’s ability to speed development of new AI models | ForkLog

AI summary

Anthropic released Claude Fable 5.2 and Mythos 5.2 for complex engineering and scientific tasks, and deliberately limited during training capabilities that could accelerate stronger AI R&D, involving software optimization for specialized compute accelerators and large language model architecture improvements; in these areas the new models are roughly at the level of Claude Opus 5.5. According to Anthropic, Fable 5.2 scored 49.1% on the FrontierCode 2.0 coding test (Xhigh compute setting), higher than Fable 5.1's 32.4%, and scored 64.9% on the RoboDojo Sim robotics test (previous version 27.6%), and once coordinated more than 1000 AI agents to build an operating system prototype including kernel, drivers and graphical interface in eight days. Shortly before the release, METR research estimated that since October 2025 the rate of improvement in Anthropic model test scores was about 1.9 times that of the earlier period, with about 80% of the acceleration coming from using AI to develop new systems.

Read originalforklog.com