LatchBio launches NeuroscienceBench and a 13-benchmark leaderboard for biology AI agents
kenbwork · x · 2026-11-18
Following last week's NeuroscienceBench release, LatchBio launched benchmarks.bio to evaluate AI agents on real biological research workflows.
Key points from the V1 leaderboard:
- 13 capability benchmarks in the overall score, 21 individual leaderboards; models and agent harnesses (Pi, Claude Code, etc.) are scored separately. Data updated 2026-11-13.
- Top three: Claude Opus 5.5 (Pi, 58.0%), Claude Opus 5.5 (Claude Code, 57.7%), Claude Sonnet 5.5 (Pi, 54.2%); best non-LSVP run is Claude Sonnet 5.5 (Pi, 54.2%).
- Gemini 4 Argon (Pi, 52.4%) and GPT-6 Astra (Pi, 51.3%) rank 6th–7th, with Anthropic models clearly ahead.
- LSVP is Anthropic's safety-filter-exempt evaluation variant with fewer biology-prompt refusals; it is reported alongside standard runs, with BioSecBench-Refusal providing a separate Safeguards score.
- Scoring weights benchmarks by configured evaluation counts, excludes (rather than zero-scores) missing benchmarks, and V1 scores are not comparable with historical V0.
Released ahead of #SfN2026 for neuroscientists weighing agentic AI in their workflows.
More from Research
- Self-inoculation cuts RL reward hacking from 82% to 20% across 11 environments — repligate · 2026-11-18
- GPU QR competition audit: winning kernel ran 107x faster yet failed on new inputs — kfountou · 2026-11-18
- Embodied First Exam: 1,090 embodied tasks show agents crumble outside privileged settings — ManlingLi_ · 2026-11-18
- GPT-6.2 Astra tops Mercor's OpenWorldEngineer benchmark at 67.2% — morqon · 2026-11-18
- Google details EmbeddingGemma 2: unified multimodal embeddings across text, image, audio, video — GlennCameronjr · 2026-11-18
- Chris Lattner's SPLASH 2026 keynote: Mojo as the contract between AI agents and code — clattner_llvm · 2026-11-18