Anthropic on Tuesday launched Claude Fable 5, a public, guardrailed model of its vaunted “Mythos” AI mannequin.
The AI startup mentioned Claude Fable 5 is extra succesful than some other mannequin it has launched to the general public, displaying excellent efficiency in software program engineering, information work, and imaginative and prescient duties, amongst different areas.
The mannequin is superior to earlier iterations in performing longer, extra complicated duties. The analytics firm Hex mentioned that Fable 5 was the primary mannequin to interrupt 90% on its core benchmark of complicated, long-running analytical duties–a 10-point soar over earlier Opus fashions. Claude Fable 5 scored 80.3% on the SWE-Bench-Professional, which measures AI fashions’ agentic coding ability, in comparison with (OpenAI) GPT-5.5’s rating of 58.6%, and (Google Deepmind) Gemini 3.1 Professional’s rating of 54.2%. The testers mentioned Fable confirmed “sturdy judgment and a focus to nuance.”
Anthropic launched the unique Mythos Preview model in April, however solely to a choose set of defensive cybersecurity execs and overseers of essential infrastructure. Mythos, the corporate mentioned, had proved to be excellent at discovering and exploiting safety weaknesses in industrial software program.
The brand new Fable 5 mannequin comes with a number of limitations, Anthropic says. It gained’t give responses to questions within the areas of cybersecurity and biology, for instance, as a substitute routing such inquiries to the less-capable Claude Opus 4.8 mannequin. This occurred in solely about 5 % of periods, nevertheless, Anthropic says.
Fable 5 was subjected to makes an attempt by each inner and exterior groups to trick it into answering questions on prohibited topic areas. None of those makes an attempt have been profitable.
As for the cybersecurity researchers who have already got entry to Mythos Preview, they’ll now get entry to a brand new mannequin referred to as Claude Mythos 5. Extra individuals and organizations will get entry via “a extra systematic trusted-access program,” Anthropic says.
This public launch of a Mythos mannequin comes simply days after Anthropic urged main AI laboratories to implement a coordinated global pause on growth of AI, warning of rising threat that fast developments in AI capabilities might outrun human management mechanisms.
Anthropic’s rival OpenAI doesn’t disagree. In a brand new blog post Monday, CEO Sam Altman (with Jakub Pachocki) wrote that the business wants a world group that may “make it potential for the world to take coordinated motion, together with slowing frontier growth when wanted, so societal resilience, security, and alignment can hold tempo.”
AI firm leaders have referred to as for extra aggressive authorities regulation on frontier mannequin growth, however the AI business foyer and the Trump Administration have labored arduous to derail binding AI security legal guidelines that require greater than voluntary compliance from AI labs.
Nonetheless, the potential safety risk of Mythos reportedly fueled disagreement amongst key White Home staffers in regards to the power of the federal government’s oversight position. It additionally helped set off the latest AI executive order, which asks AI labs to provide the federal government a month of advance testing earlier than they launch sure new frontier fashions.
Claude Fable 5 is accessible to builders by way of Anthropic’s API.
