Anthropic Releases Claude Fable 5.2 and Claude Mythos 5.2
Anthropic releases Claude Fable 5.2 and Claude Mythos 5.2. Both are based on the same underlying model, target the hardest technical work, and offer improvements in coding, reasoning, and vision. Fable comes with stronger cybersecurity and biosecurity guardrails and is now available on paid plans and major cloud platforms; Mythos requires applying through the life sciences or cybersecurity verification program. On FrontierCode 2.0, Fable 5.2's High tier scores about 45.3%, roughly on par with Opus 5.5's top score of 45.5% while costing about half per task, and its Xhigh tier scores 49.1%.
Why read this
SelectedAI score 82
The original text provides specific comparisons between the new models and Opus 5.5 on cost, long-horizon tasks, and robotics benchmarks, making it easier to judge the applicable boundaries of this tier of models.
Topics
View original postIntroducing Claude Fable 5.2 and Claude Mythos 5.2, designed to tackle the most daunting technical challenges. They’re capable collaborators on ambitious projects in many fields, and show improvements in areas like coding, reasoning, and vision. https://t.co/jP0AutLxXb
View original
Introducing Claude Fable 5.2 and Claude Mythos 5.2, designed to tackle the most daunting technical challenges. They’re capable collaborators on ambitious projects in many fields, and show improvements in areas like coding, reasoning, and vision. https://t.co/jP0AutLxXb
Body · AI translation
Anthropic releases Claude Fable 5.2 and Claude Mythos 5.2, aimed at the toughest technical work, with improvements in coding, reasoning, and vision. btw... still didn't get GPT 6.2 in the end...
Claude Fable 5.2 and Claude Mythos 5.2 are the same underlying model. Fable comes with stronger cybersecurity and biosecurity guardrails and is already available on paid plans and major cloud platforms; Mythos requires applying through a life sciences or cybersecurity verification program and is aimed at beneficial biomedical or cybersecurity research. For most everyday coding and knowledge work, the official recommendation is still to start with Opus 5.5, and switch to this tier for complex, open-ended long-horizon tasks.
https://t.co/ca4BmhaHwS
On alignment, most metrics are no lower than recent models, and pre-release external evaluations were conducted by the UK AI Safety Institute, Frontier Design, METR, and others. On capability, it is higher than Opus 5.5 on almost every evaluation dimension. On FrontierCode 2.0, Fable 5.2 at the High setting scores about 45.3%, roughly on par with Opus 5.5's top score of 45.5%, at about half the cost per task; at the Xhigh setting it scores 49.1%, the highest among tested and released models. The application default is the Medium setting.
Demonstration results: about 1,000 Fable 5.2 agents formalized Fermat's Last Theorem in 30 hours; under the same setup, Opus 5.5 completed about 26% in 18 hours and made almost no further progress after 30 hours. Mythos accurately modeled 28 of the 34 gene therapy capsids it attempted. On the simulated dual-arm benchmark RoboDojo, the internal top setting scored 64.9%, higher than Opus 5.5's 39.4% and Fable 5.1's 27.6%.