What Claude Mythos 5 Means for Science: Drug Design, Genomics, and Trusted Access
Claude Mythos 5 accelerated drug design 10x in internal tests and did autonomous genomics across 138 species. Here is what the science results mean.
TL;DR
Claude Mythos 5 - the lifted-safeguards counterpart to Claude Fable 5, announced June 9, 2026 - posted striking science results: internal protein-design experts accelerated drug design roughly 10x, 9 of 14 protein targets yielded strong drug-design candidates, and blind comparisons preferred its molecular biology hypotheses about 80 percent of the time over Opus-class models. Access is restricted to vetted organizations, with select biomedical researchers joining later through a trusted access program.
The Drug Design Results
The most consequential numbers come from protein and drug design. Working with Mythos 5, Anthropic's internal protein-design experts accelerated their drug design work by roughly 10x. Across 14 protein targets attempted, 9 yielded strong drug-design candidates - a hit rate that, sustained at scale, would meaningfully change early-stage pharmaceutical timelines.
Hypothesis quality held up under blind evaluation: molecular biologists preferred Mythos 5's hypotheses about 80 percent of the time when compared against Opus-class models, without knowing which model produced which. And the work has begun touching ground truth - one hypothesis about an E. coli protein was independently corroborated.
Autonomous Genomics
Beyond assisted design, Mythos 5 demonstrated autonomous science. In a genomics project it assembled single-cell data across 138 species and trained a custom machine learning model 100x smaller than the comparison system - and that smaller model outperformed a publication in a Science journal. The combination is notable: data assembly, method design, and model training executed end to end, with a result competitive with peer-reviewed work.
Who Gets Access, and Why It Is Restricted
Mythos 5 is the same underlying model as the publicly available Fable 5, but with safeguards lifted in some areas - and it has the strongest cybersecurity capabilities of any model in the world. Access therefore runs through controlled channels:
- Project Glasswing, in collaboration with the US government, serving cyberdefenders and infrastructure providers - upgrading the earlier Claude Mythos Preview and expanding to roughly 150 new organizations across more than 15 countries
- A trusted access program for select biomedical researchers, coming later
The model that designs drug candidates faster also lowers barriers in dual-use biology, which is exactly why general users get Fable 5 instead: the same intelligence with biology and chemistry classifiers in place, safeguards triggering in under 5 percent of sessions, and over 1,000 hours of external red-teaming behind it.
What Researchers Should Do Now
Biomedical researchers should watch for the trusted access program's criteria. Everyone else in science still gets most of the benefit through Fable 5 - state-of-the-art extraction of precise numbers from scientific figures, millions of tokens of focused context for literature-scale work, and frontier reasoning - free on paid Claude plans from June 9 to 22, then via usage credits and the Claude API.