LabRobotBench: best AI agent scores just 46% at driving a robotic wet lab
danielmckinn0n · x · 2026-10-21
GamowLabs, which aims to understand the clinical impact of 9B+ possible human genome mutations for rare disease, unveiled LabRobotBench — testing whether frontier AI agents can run a robotic wet lab: operating equipment, managing timing, and recovering interrupted experiments, simulated on their real Boulder, CO setup.
- The best model tested, GPT 6.1 Sol, scores only 46%
- Their argument: computational models like AlphaGenome, Evo2 and ESM3 help, but the mission requires large-scale experimental variant-effect data, which only robotic labs can produce at scale — so they're building frameworks for AI to drive lab robots directly
- Improving on the benchmark implies better robotic control, spatial awareness, and biological reasoning
More from Embodied
- fal starts fine-tuning a physical robot to draw, results next week — OdinLovis · 2026-10-23
- PEARL demo: an LLM agent that checks experiment results, revises plans, and re-tests — KevinKaichuang · 2026-10-23
- CU Boulder's ATLAS Institute Opens Tenure-Track Robotics Professorship — ryosuzk · 2026-10-23
- MiTaS multi-resolution tactile imitation learning hits 80% success vs 35.5% vision-only baseline — Jan_R_Peters · 2026-10-23
- Princeton's Action Self-Distillation Teaches Robots Language Navigation with Zero Human Labels — shahdhruv_ · 2026-10-23
- Coding agents like Codex can already handle much of real-to-sim for robot manipulation — zhenjun_zhao · 2026-10-23