Skip to content

Benchmark

ScienceQA (val)

Science benchmarks measure how well models answer science questions. Scores are percentages.

Models measured
34
Scores recorded
34
Sources
1
IQ weight
1.18

Score distribution

83.3%
Mean
87.1%
Median
19.3%
Std dev
71.6%
10th percentile
97.9%
90th percentile
1.3%
Min
1.3%99.2%
The band covers the middle 80% of scores. The line marks the median.

This benchmark has a weight of 1.18 for chat models, measured over 34 models. The weight is how much it counts toward llminfo IQ. We measure it from the data, not from the raw scores.

Leaderboard

#ModelScore
1internvl3-38b· OpenVLM Leaderboard data99.2%
2GLM 4 Vision· OpenVLM Leaderboard data98.6%
3InternVL3 78B· OpenVLM Leaderboard data98.5%
4InternVL2 26B· OpenVLM Leaderboard data97.9%
5internvl3-8b· OpenVLM Leaderboard data97.9%
6GLM 4 Vision Plus· OpenVLM Leaderboard data96.8%
7Phi 4 Multimodal Instruct· OpenVLM Leaderboard data96.3%
8InternVL2 4B· OpenVLM Leaderboard data96%
9MiniCPM-V 2.6· OpenVLM Leaderboard data96%
10Hunyuan Standard Vision· OpenVLM Leaderboard data95.5%
11GPT 4.1· OpenVLM Leaderboard data92.6%
12Qwen2-VL-72B· OpenVLM Leaderboard data91.6%
13Qwen2.5 VL 72B Instruct· OpenVLM Leaderboard data91.3%
14Grok 2 Vision 1212· OpenVLM Leaderboard data90.3%
15Phi-3.5 Vision Instruct· OpenVLM Leaderboard data90.3%
16Qwen2.5-VL 7B Instruct· OpenVLM Leaderboard data88.9%
17Gemma 3 27B IT· OpenVLM Leaderboard data87.1%
18Pixtral 12B· OpenVLM Leaderboard data86.6%
19Yi Vision· OpenVLM Leaderboard data86.5%
20Gemini 1.5 Pro· OpenVLM Leaderboard data85.9%
21Gemma 3 12B IT· OpenVLM Leaderboard data85.2%
22Llama 3.2 90B Vision Instruct· OpenVLM Leaderboard data83.5%
23Gemini 1.5 Flash· OpenVLM Leaderboard data82.9%
24GPT-4.1 mini· OpenVLM Leaderboard data81.7%
25Llama 3.2 11B Vision Instruct· OpenVLM Leaderboard data80.9%
26gemini-1.0-pro· OpenVLM Leaderboard data80.1%
27Qwen-VL Max· OpenVLM Leaderboard data79.8%
28qwen2.5-vl-3b-instruct· OpenVLM Leaderboard data79.3%
29GPT-4.1 nano· OpenVLM Leaderboard data76.8%
30Gemma 3 4B· OpenVLM Leaderboard data76%
31Qwen-VL Plus· OpenVLM Leaderboard data71.6%
32Molmo 7B-D· OpenVLM Leaderboard data55.6%
33kosmos-2· OpenVLM Leaderboard data33.1%
34Moonshot V1 8K· OpenVLM Leaderboard data1.3%

Data credits: OpenVLM Leaderboard (OpenCompass / VLMEvalKit), Apache-2.0 (licence)