GeoMMBench: leaderboard

Expert-written multimodal geoscience and remote sensing questions on the GeoMMBench (2026) test set, spanning remote sensing, photogrammetry, GIS and GNSS across optical, SAR, LiDAR, elevation, spectral and thermal data, answered zero-shot. Micro-averaged accuracy in percent; higher is better.

Metric: Micro-averaged accuracy (%) on all questions of the GeoMMBench test set (expert-written multimodal geoscience and remote sensing multiple-choice and numeric questions, zero-shot, LMMs-Eval, rule-based answer extraction, invalid outputs counted wrong); higher is better. Source: arxiv.org. Saturation forecast: Around February 2027. 35 models tracked. Updated 7 Oct 2026.

Top models

#ModelScoreOverall rank
1Gemini 1.5 Pro70.7#1063
2Gemini 2.0 Flash70.1#723
3GPT-4o68.5#751
4Qwen 3 VL 30B66.7#561
5O165.8#403
6Qwen 3 VL 8B65.8#917
7Qwen 3 VL 4B63.2#1151
8Qwen 2.5 VL 7B62.5#1862
9GPT-4V61#1470
10Qwen 2 VL 7B59.8#2275
11Qwen 2.5 VL 3B58.1#2639
12LLaVA-OneVision-7B57.3#2981
13InternVL2.5-8B56.8#2370
14Phi-4 Multimodal Instruct55.3#4154
15LLaVA-NeXT-34B54.2#3443

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

Interactive version: theaggregate.ai/benchmark?slug=geommbench · How It Works · Data refreshed daily, snapshot 2026-11-21.