Anthropic Launches Fable 5.2 and Mythos 5.2 with Science and Coding Upgrades
New models achieve notable AAV capsid modeling capabilities and 22% performance gain on scientific benchmarks while reducing costs by 20%.
Anthropic has unveiled two advanced large language models that promise to accelerate biomedical research and gene therapy development. Claude Fable 5.2 and Claude Mythos 5.2, released November 9, deliver significant improvements in scientific research and coding capabilities, with the models achieving a 22% performance gain on scientific benchmarks compared to their predecessors.
A notable advancement for healthcare applications is the models' ability to accurately model 28 of 34 adeno-associated virus (AAV) capsid structures—complex assemblies comprising dozens of proteins each. AAV vectors serve as the primary delivery mechanism for approved gene therapies treating conditions including spinal muscular atrophy and hemophilia, making improved capsid modeling capabilities directly relevant to therapeutic development pipelines.
Outperforming Established Protein Modeling Technology
The new models demonstrated superiority over Google DeepMind's AlphaFold2 by modeling entire multi-protein capsid structures rather than individual proteins alone. The creators of AlphaFold2 won one half of the 2024 Nobel Prize in Chemistry. On the Terminal-Bench-Science 0.1 benchmark, Fable 5.2 scored 74.6%, representing the 22% improvement over its predecessor Fable 5.1.
Beyond protein modeling, the models showed dramatic efficiency gains in mathematical proof generation, reducing computer-verifiable proof creation from 11 days to 30 hours using 1,000 parallel agents. In a separate demonstration, the system built a functional laptop operating system with network access in eight days using the same parallel agent approach.
Coding performance improved by more than 13% over the previous generation on FrontierCode 2.0 and FrontierSWE 2.0 benchmarks, with Fable 5.2 outperforming its predecessor by over 10% across most of 14 AI benchmarks tested.
Two-Tier Access Model Balances Safety and Innovation
Anthropic has implemented a differentiated access strategy for the new models. While Fable 5.2 and Mythos 5.2 share identical core capabilities, they differ in guardrail implementation. Mythos 5.2 features relaxed guardrails for biology and cybersecurity research but is available only through limited-access programs for trusted researchers.
Fable 5.2, the publicly available version, blocks certain sensitive use cases in biology and cybersecurity but routes those queries to less capable models—specifically Opus 4.8 or Opus 5—when needed. According to Anthropic, Fable 5.2 is 55% less likely to erroneously block safe prompts compared to Fable 5.1, addressing a common friction point in AI-assisted research workflows.
Pricing and Availability
The models are priced at $8 per million input tokens and $40 per million output tokens, representing a 20% cost reduction compared to the previous generation. For applications requiring faster response times, Anthropic offers a high-performance mode that generates responses 2.5 times faster at double the standard price.
Anthropic positions the new models for highly complicated, open-ended technical tasks, recommending its Opus 5.5 model, released in July 2026, for day-to-day workflows. Users can also combine the models by having Opus 5.5 ask Fable 5.2 or Mythos 5.2 for advice when it encounters a bottleneck.
Sources
Google Search
Follow Health AI Central on Google
Choose Health AI Central as a preferred source, and Google is more likely to show our reporting in Top Stories.
Reporting & accountability
How we maintain accuracy
Health AI Central separates reported facts, analysis, and public-source data. Readers can review our standards, request a correction, or consult the relevant research methodology before relying on a claim.