HM-Bench (Report): 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 report-only input: a structured text report of spectral and spatial statistics computed from the 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 report-only input: a structured text report of spectral and spatial statistics computed from the 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. 16 models tracked. Updated 7 Oct 2026.

Top models

#ModelScoreOverall rank
1InternVL3.5-14B39.52#1504
2GeoLLaVA-8K38.76
3GLM-4.6V-Flash37.72#1040
4Kimi-VL-A3B-Instruct37.57#2253
5GPT-5.4 Mini37.45#406
6InternVL3-14B37.26#1214
7InternVL3-9B37.18#1728
8Llama-3.1-Nemotron-Nano-VL-8B-V137
9LLaVA-Next-8B36.71#2046
10Claude Sonnet 4.636.57#173
11InternVL2-8B36.17#3227
12Qwen 3 VL 4B35.93#1151
13GeoChat34.06#3155
14InternVL2.5-4B32.3#2549
15Grok 431.68#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-report · How It Works · Data refreshed daily, snapshot 2026-11-21.