CellType's BioSimBench tests frontier LLMs on real lab experiments; 27B model beats frontier models
SimplyAnnisa · x · 2026-11-04
CellType released BioSimBench, a benchmark evaluating how well LLMs simulate real biological behavior using actual lab measurements across drugs, gene knockdowns, and immune signals. Even the best frontier models get only part of the cell response right. The company's 27B model, trained via BioSimGym, beats frontier models on every task, built with NVIDIA and NeMo RL — a promising step for AI in biology.
More from Research
- Adversarial speculative decoding boosts long-horizon agent accuracy, paper finds — SubconsciousDev · 2026-11-05
- 2-person lab trains 2B text-to-image model from scratch for $20K with Pyramid-JiT — generativist · 2026-11-05
- Axiom ships 16M-line AI formalization of the Stacks Project, countering 'AI code is slop' — Scobleizer · 2026-11-05
- 2-person lab releases 2B pixel-space text-to-image model trained for ~$20K — liuziwei7 · 2026-11-05
- Frontier training without megacampuses: MoE model trained across <20MW clusters over 25Gbps — _arohan_ · 2026-11-05
- GRETA platform screens 1,000 FDA drugs, finds statins may reverse IPF signature — danielmckinn0n · 2026-11-05