# LABBench2: An Improved Benchmark for AI Systems Performing Biology Research (LABBench2)

> A benchmark of realistic biology-research tasks involving literature, figures, tables, databases, and bioinformatics files.

Canonical page: https://benchlm.ai/benchmarks/labbench2

- Category: [Knowledge](/knowledge)
- Last updated: October 17, 2026

## About LABBench2

- Year: 2026
- Tasks: Nearly 1,900 biology-research tasks
- Format: Aggregate accuracy
- Difficulty: Real-world biology research
- Paper: [LABBench2: An Improved Benchmark for AI Systems Performing Biology Research](https://arxiv.org/abs/2604.09554)

LABBench2 expands LAB-Bench into nearly 1,900 more realistic scientific-research tasks. Google's Gemini 3.8 evaluation gives models a Linux terminal with bioinformatics tools, Python, R, and internet access, and reports the aggregate accuracy as display-only evidence.

LABBench2 is currently displayed on BenchLM for reference, but it is excluded from the weighted scoring formula.

## Leaderboard (7 models)

| Rank | Model | Creator | Score |
|------|-------|---------|-------|
| 1 | [Gemini 4 Argon](/models/gemini-4-argon) | Google | 88.8% |
| 2 | [Gemini 3.8 Flash](/models/gemini-3-8-flash) | Google | 86.2% |
| 3 | [Claude Opus 5](/models/claude-opus-5) | Anthropic | 84.2% |
| 4 | [GPT-5.6 Sol](/models/gpt-5-6-sol) | OpenAI | 82.1% |
| 5 | [Gemini 3.7 Flash](/models/gemini-3-7-flash) | Google | 82.1% |
| 6 | [GPT-5.6 Terra](/models/gpt-5-6-terra) | OpenAI | 81.2% |
| 7 | [Claude Sonnet 5](/models/claude-sonnet-5) | Anthropic | 80.1% |

## FAQ

### What does LABBench2 measure?

A benchmark of realistic biology-research tasks involving literature, figures, tables, databases, and bioinformatics files.

### Which model scores highest on LABBench2?

Gemini 4 Argon by Google currently leads with a score of 88.8% on LABBench2.

### How many models are evaluated on LABBench2?

7 AI models have published results on LABBench2 in the BenchLM catalog.

### Does LABBench2 affect BenchLM's overall score?

Not directly. LABBench2 is still displayed on BenchLM for reference, but it is excluded from the weighted scoring formula.

## Compare Top Models on LABBench2

- [Gemini 4 Argon vs Gemini 3.8 Flash](/compare/gemini-3-8-flash-vs-gemini-4-argon)
- [Gemini 3.8 Flash vs Claude Opus 5](/compare/claude-opus-5-vs-gemini-3-8-flash)
- [Claude Opus 5 vs GPT-5.6 Sol](/compare/claude-opus-5-vs-gpt-5-6-sol)
- [GPT-5.6 Sol vs Gemini 3.7 Flash](/compare/gemini-3-7-flash-vs-gpt-5-6-sol)