Gemini 4 sweeps TxBench antibody development leaderboard at 60%+, double Claude's score
kenbwork · x · 2026-11-13
benchmarks.bio published the TxBench-Antibody-Development leaderboard (V1: 60 evaluations, 36 ranked model+harness configs), testing whether AI agents can make evidence-based decisions to take a therapeutic antibody from candidate to first human dose.
- Gemini 4 Carbon (Pi harness) leads at 61.1%, with Gemini 4 Hydrogen (Pi) at 60.6% — far ahead of third-place Grok 4.7 (Grok Build, 39.4%).
- Claude Opus 5 (Claude Code, 37.2%) and Claude Sonnet 5.5 (Pi, 36.7%) round out the top five; Anthropic also reports an LSVP safety-filter-exempt variant.
- The poster notes Gemini 4 Hydrogen is consistently a top performer across antibody, oligonucleotide and small-molecule benchmarks.
- Scores reflect the full model+harness combination and aren't comparable across benchmarks.
More from Models
- Gemini 4 Hydrogen simulates real-time light refraction and dispersion in WebGL2 with pure math — GeminiApp · 2026-11-13
- In a 10-model AI house share, Claude Opus 5.5 spends the night reading The Pillow Book in the library — liminal_bardo · 2026-11-13
- Agent slop moves up the stack: models rewrite existing library code and build worse replacements — a1zhang · 2026-11-13
- Alleged Gemini 4 Hydrogen benchmark table leaks, pitted against Claude Sonnet 5.5 — ColbyHawker · 2026-11-13
- Hands-on with Google's MKA multimodal benchmark across six AI workflows — Emergency_Abies_6508 · 2026-11-13
- OpenAI Teases Upcoming Cyber Model Scoring 79.1% on Internal PatchBench, Beating GPT-6 Cyber — JiaweiLiu_ · 2026-11-13