HM-Bench (RGB): leaderboard

HM-Bench (2026) tests multimodal LLMs on hyperspectral image understanding with 19,337 multiple-choice questions over 2,178 samples in 13 perception and reasoning tasks, given RGB-only input: a pseudo-RGB rendering of the hyperspectral cube. Accuracy (%), zero-shot; higher is better.

Metric: Average accuracy (%) over all tasks of HM-Bench's 19,337 hyperspectral multiple-choice questions (2,178 samples, 13 task categories) with RGB-only input: a pseudo-RGB rendering of the hyperspectral cube; zero-shot, one option letter as output, temperature 0, at most 64 new tokens; higher is better. Source: arxiv.org. Saturation forecast: Around 2033. 17 models tracked. Updated 7 Oct 2026.

Top models

#ModelScoreOverall rank
1GLM-4.6V-Flash44.46#1049
2InternVL3.5-14B42.88#1519
3InternVL3-14B42.48#1227
4InternVL3-9B41.74#1746
5Claude Sonnet 4.640.95#179
6Kimi-VL-A3B-Instruct40.24#2272
7Qwen 3 VL 4B40.08#1163
8GPT-5.4 Mini39.88#409
9GeoLLaVA-8K39.27
10Qwen 2.5 VL 7B Instruct38.19#1919
11Llama-3.1-Nemotron-Nano-VL-8B-V137.54
12LLaVA-Next-8B37.39#2058
13GeoChat37.03#3175
14InternVL2.5-4B36.74#2573
15InternVL2-8B36.63#3268

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

Interactive version: theaggregate.ai/benchmark?slug=hm-bench-rgb · How It Works · Data refreshed daily, snapshot 2026-11-22.