MicroVQA - Experiment Proposal: leaderboard

Accuracy (%) on the MicroVQA questions of the experiment proposal task: choosing the next experiment to test a hypothesis from microscopy images. Each model answers every question of this task once with the same prompt as the headline board.

Metric: Accuracy (%). Source: jmhb0.github.io. Saturation forecast: Estimated already saturated. 14 models tracked. Updated 17 Nov 2026.

Top models

#ModelScoreOverall rank
1O1 (2024-12-17)53#314
2Claude 3.5 Haiku (20241022)51.7#1589
3Qwen 2 VL 7B Instruct49.6#2594
4Gemini 1.5 Flash-8B (001)49.1#2311
5Grok 2 (1212)48.7#1082
6pixtral-large-241148.7#1284
7Qwen 2 VL 72B Instruct47.8#1382
8VILA1.5-40B47.4#3238
9GPT-4o Mini (2024-07-18)47#1494
10Llama 3.1 Nemotron 70B Instruct44.8#1471
11Pixtral-12B-240944.8#2537
12VILA1.5-13B40.9#7029
13Llama 3.2 90B Vision Instruct38.7#1791
14Llama 3.2 11B Instruct28.7#5226

No result here: #1 Claude Fable 5.2, #2 GPT-6.2 Astra, #3 Gemini 4 Carbon.

Interactive version: theaggregate.ai/benchmark?slug=microvqa-experiment-proposal · How It Works · Data refreshed daily, snapshot 2026-11-17.