AlphaGenome Benchmarks: Better Than Prior SOTA Overall, But Not a Giant Leap in Most Cases

anshulkundaje · x · 2026-09-18

The original AlphaGenome paper collates benchmark evaluations against perturbations, synthetic libraries, silver-standard causal variant sets and more. It is generally decently better than previous SOTA, though not a giant leap in many cases.

Related event: Stanford expert defends AlphaGenome variant scores in detailed thread, rebutting "AI slop" criticism(20 posts)→

Original post →

More from Research

Research channel →