Claude Fable 5.2 and Mythos 5.2: What can they actually do, and how independently can they work?
Claude Fable 5.2 and Mythos 5.2 can operate unattended for days to solve complex coding tasks. While Fable is public, the highly capable Mythos model is strictly restricted to vetted cybersecurity and life sciences research.
A New AI Era
Anthropic has launched Claude Fable 5.2 and Mythos 5.2, setting a new benchmark for autonomous artificial intelligence. These frontier models are designed to tackle highly complex technical tasks across coding, mathematics, and advanced scientific research.
Running Unattended
Unlike standard AI that requires constant human prompting, Fable 5.2 can operate completely independently for hours or even days. The model interprets ambiguous instructions, iterates on solutions, verifies its own output, and automatically corrects errors as it works.
Real-World Testing
During pre-release evaluations, these models successfully completed massive software migrations and designed complex electronic devices from scratch. In one test, a team of 1,000 AI agents even solved a formal proof of Fermat's Last Theorem within 30 hours.
The Mythos Safeguard
While Fable 5.2 is broadly accessible to the public, Mythos 5.2 remains strictly locked behind corporate verification programmes. Mythos is specifically reserved for vetted organisations conducting high-risk operations like penetration testing, cybersecurity engineering, and biomolecular design.
Faster and Cheaper
The new systems generate outputs up to 30 per cent faster than their predecessors while consuming significantly less computing power. Anthropic has priced the models at USD 8 per million input tokens and USD 40 per million output tokens.
Finding Vulnerabilities
Mythos 5.2 is Anthropic's most capable model for digital security engineering to date. During internal testing, it successfully discovered 10.2 per cent more critical software vulnerabilities than the previous Opus 5.5 model.
Halting Illicit Scraping
To protect its proprietary technology, Fable 5.2 features the strongest corporate protections yet against illicit data distillation. The model actively blocks attempts to harvest its data, reducing successful scraping by 90 to 99 per cent.
)
)
)
)
)