CellType claims 27B biology model beats all 12 frontier models at predicting cell behavior
alifcoder · x · 2026-11-05
CellType released BioSimBench (built on NVIDIA), testing frontier models on real lab experiments: predicting how cells respond to drugs, gene knockdowns, and immune signals. Even the best models only get part of the response right.
Per the relay, CellType's 27B model trained on measured lab biology beats all 12 frontier models tested, and leads on six benchmarks from outside labs — margins range from nearly 2x on drug response to 0.121 vs 0.119 on one test.
Other labs will want to verify; if it holds up, it marks a milestone for AI that predicts biology instead of describing it.
More from AGI Musings
- Kevin Roose interviewed 150 AI insiders for The AGI Chronicles, fact-checked with a council of Claudes — every · 2026-11-05
- AutoDNA: AI agents run multi-step molecular biology experiments with minimal human input — james_y_zou · 2026-11-05
- Amazon blocks Meta's Muse shopping agent over ad revenue threat — XIFAQ · 2026-11-05
- bfl_ai co-founder explains world models: intelligence learned from natural signals — shiringhaffary · 2026-11-05
- Kai-Fu Lee in TIME: why almost every corporate AI transformation fails — kaifulee · 2026-11-05
- Lance Fortnow: AI now finds theorem proofs on its own, Optiland arrived five years early — fortnow · 2026-11-05