Skip to content
ynetglobal·Indexed by this site · Originally published· AI updates →SelectedAttention48

Anthropic releases two AI models so capable that its own developers are concerned

Anthropic launches two AI models powerful enough to worry their own developers

AI summary

Anthropic released Claude Fable 5.2 and Claude Mythos 5.2 on Monday, saying they can complete engineering and scientific tasks previously considered beyond AI capabilities, with output up to 30% faster than the previous generation and API pricing reduced to $8 per million input tokens and $40 per million output tokens. The two models are based on the same underlying model; Fable 5.2 targets a broad user base and has stricter restrictions on cybersecurity and biological requests, while Mythos 5.2 opens up more sensitive-domain capabilities and is limited to vetted institutions and researchers. Demonstrated results include more than 1000 sub-agents collaborating for 8 days to write a bootable prototype operating system, and about 1000 agents completing a Lean formalization of the proof of Fermat's Last Theorem in about 30 hours. METR reported last week that the rate of progress of Anthropic models on some tasks since October 2025 has been about 1.9 times the previous rate, and believes much of the acceleration may be related to the company's expanded use of AI in its own R&D.

Why it was selected

Beyond the release information, the report provides test data on the model bypassing the sandbox and METR's speedup evaluation, allowing this release to be understood in the context of capability acceleration and safety challenges coexisting.

Read originalynetnews.com