BIABench: leaderboard
BIABench (2026) — 16 end-to-end bioimage-analysis tasks reconstructed from published biological studies (segmentation, counting, tracking, colocalization, stitching and 3D quantification across histology, fluorescence, light-sheet and time-lapse microscopy). An agent receives raw images and a brief written instruction, and its deliverable is scored against the study's ground truth and expert rubrics; mean outcome over three runs per task.
Metric: Outcome score (0-100): mean over 16 end-to-end bioimage-analysis tasks of the per-task mean of three runs, scored against each source study's ground truth (brief instruction, DeepSeek Harness, medium effort). Source: biabench.github.io. Saturation forecast: Around March 2028. 6 models tracked. Updated 1 Oct 2026.
Top models
| # | Model | Setting | Score | Overall rank |
|---|---|---|---|---|
| 1 | Claude Opus 5 | Medium | 65.4 | #20 |
| 2 | GPT-5.6 Sol | Medium | 58.3 | #30 |
| 3 | Kimi K2.6 | Medium | 56.9 | #181 |
| 4 | DeepSeek V4 Pro | Medium | 52.7 | #191 |
| 5 | DeepSeek V4 Flash | Medium | 52.1 | #259 |
| 6 | GLM-5.1 | Medium | 50.6 | #186 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=biabench · How It Works · Data refreshed daily, snapshot 2026-11-16.