LatchBio's antibody discovery benchmark: Opus and Gemini lead, OpenAI models underperform
kenbwork · x · 2026-09-03
LatchBio introduces an Antibody Discovery Benchmark testing whether AI agents can make scientific decisions across stages of therapeutic antibody discovery.
- 100 evaluations across 10 areas drawn from concrete drug programs: target and modality selection, assay design, binder discovery, binding characterization, cellular pharmacology, antibody engineering, and preclinical candidate de-risking.
- Discovery is multiparameter and context-dependent: affinity must be weighed against specificity, stability, solubility, expression, and biological activity, and every measurement interpreted within its experimental system.
Early results: Claude Opus and Gemini do well; OpenAI models do poorly overall — a surprising gap.
Related event: LatchBio's Antibody Benchmark: Opus and Gemini Lead, OpenAI Lags(4 posts)→
More from Research
- CBAI opens Fall AI Safety fellowship: $15k stipend, 10 weeks in Boston — benno_krojer · 2026-09-03
- What 12 million empirical research results can teach us — RexDouglass · 2026-09-03
- GamowLabs to launch RareBench challenge: can you tell synthetic genomes from real ones? — danielmckinn0n · 2026-09-03
- NEJM: Single-Dose In-Vivo B-Cell Depletion Helps All 16 Refractory Autoimmune Patients — EricTopol · 2026-09-03
- Multi-Teacher On-Policy Distillation emerges as 2026 post-training paradigm used in frontier models — cwolferesearch · 2026-09-03
- Protein folding models have always used recycling — effectively recurrent depth with stop grad — amyxlu · 2026-09-03