HM-Bench (PCA): 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 PCA-only input: a false-colour image of the cube's leading principal components. 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 PCA-only input: a false-colour image of the cube's leading principal components; zero-shot, one option letter as output, temperature 0, at most 64 new tokens; higher is better. Source: arxiv.org. Saturation forecast: Around 2032. 16 models tracked. Updated 7 Oct 2026.

Top models

#ModelScoreOverall rank
1InternVL3.5-14B43.08#1504
2GPT-5.4 Mini42.35#406
3InternVL3-9B42.14#1728
4GLM-4.6V-Flash42.06#1040
5Qwen 3 VL 4B40.96#1151
6Claude Sonnet 4.639.48#173
7LLaVA-Next-8B39.03#2046
8Kimi-VL-A3B-Instruct38.93#2253
9Llama-3.1-Nemotron-Nano-VL-8B-V138.67
10InternVL2-8B38.09#3227
11InternVL3-14B38.09#1214
12GeoLLaVA-8K37.79
13GeoChat35.27#3155
14InternVL2.5-4B34.83#2549
15Grok 432.78#319

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

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