GeoMMBench - Perception: leaderboard

The perception questions (scene, object and sensor-level recognition) of GeoMMBench (2026), an expert-written multimodal geoscience and remote sensing question benchmark answered zero-shot (test set). Micro-averaged accuracy in percent; higher is better.

Metric: Micro-averaged accuracy (%) on the perception questions (scene, object and sensor-level recognition and fine-grained interpretation) 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 2029. 35 models tracked. Updated 7 Oct 2026.

Top models

#ModelScoreOverall rank
1Gemini 1.5 Pro68.2#1063
2Gemini 2.0 Flash66.2#723
3GPT-4o63.9#751
4Qwen 2 VL 7B63.5#2275
5O163.5#403
6Qwen 3 VL 8B62.5#917
7Qwen 3 VL 30B61.2#561
8Qwen 2.5 VL 7B59.2#1862
9Qwen 3 VL 4B58.9#1151
10GPT-4V58.5#1470
11Qwen 2.5 VL 3B57.9#2639
12CogVLM257.9
13LLaVA-OneVision-7B57.5#2981
14InternVL2.5-8B55.2#2370
15Phi-4 Multimodal Instruct53.2#4154

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

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