AI
BIABench: Evaluating AI agents on real-world bioimage analysis tasks
Publication: Hugging Face Daily Papers
Author: Not available in collected metadata
Source published: 27/09/2026, 20:00:00 UTC
Briefing created: 02/10/2026, 09:27:23 UTC
Based on a feed excerpt; full article content has not been acquired. Source evidence retrieved 01/10/2026, 18:18:24 UTC.
Briefings
BIABench is a benchmark of 16 tasks evaluating AI agents on real-world bioimage analysis, covering H&E histology to single-molecule microscopy. Tasks span eleven subtasks and modalities. Each submission receives outcome and process scores comparing outputs to ground truth. Evaluation used general-purpose and biology-specific agents across multiple language models. Routine 2D tasks were solved well, but agents failed on tasks adding 3D or time dimensions, scoring below 0.19. Biological specialization, stronger models, and expert instructions did not close this gap. Agents were unreliable, with scores varying between runs and lacking ground truth preventing correct/wrong identification.
Key points
- BIABench evaluates 16 tasks using 11 subtasks across eleven modalities from H&E histology to single-molecule localization microscopy.
- Agents failed on tasks adding 3D or time dimensions, scoring below 0.19, while 2D tasks were solved well.
Context and limitations
Scores vary between repeated runs of the same agent and cannot distinguish correct from wrong runs without ground truth. Coverage is limited to tasks with ground truth; no agent scored above 0.19 on 3D or time tasks.
Generated by Bittrees-hosted qwen35-2b sha256:aaf42c8b7c3c on 02/10/2026, 09:27:23 UTC. Based on publisher-provided feed evidence; may not cover the full original article. Check the source for details.
View archived briefing on IPFS
bafybeifgqqq2zqz7ognnhhpgt25g5irprxp5suvlnosmsrthbkfddh22ru