Skip to content

Benchmark

ScienceQA (test)

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

Models measured
34
Scores recorded
34
Sources
1
IQ weight
1.14

Score distribution

83.9%
Mean
88.5%
Median
19.2%
Std dev
73.4%
10th percentile
97.4%
90th percentile
1.2%
Min
1.2%99.3%
The band covers the middle 80% of scores. The line marks the median.

This benchmark has a weight of 1.14 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 data99.3%
2InternVL3 78B· OpenVLM Leaderboard data98.7%
3internvl3-8b· OpenVLM Leaderboard data97.9%
4GLM 4 Vision· OpenVLM Leaderboard data97.4%
5InternVL2 26B· OpenVLM Leaderboard data97.4%
6Phi 4 Multimodal Instruct· OpenVLM Leaderboard data97.3%
7GLM 4 Vision Plus· OpenVLM Leaderboard data97.1%
8MiniCPM-V 2.6· OpenVLM Leaderboard data96.7%
9Hunyuan Standard Vision· OpenVLM Leaderboard data96.4%
10InternVL2 4B· OpenVLM Leaderboard data96.3%
11GPT 4.1· OpenVLM Leaderboard data92.8%
12Qwen2.5 VL 72B Instruct· OpenVLM Leaderboard data91.4%
13Qwen2-VL-72B· OpenVLM Leaderboard data91.2%
14Grok 2 Vision 1212· OpenVLM Leaderboard data90.4%
15Qwen2.5-VL 7B Instruct· OpenVLM Leaderboard data89%
16Phi-3.5 Vision Instruct· OpenVLM Leaderboard data88.9%
17Yi Vision· OpenVLM Leaderboard data88.5%
18Gemma 3 27B IT· OpenVLM Leaderboard data87.7%
19Pixtral 12B· OpenVLM Leaderboard data87.2%
20Llama 3.2 90B Vision Instruct· OpenVLM Leaderboard data87.1%
21Gemini 1.5 Pro· OpenVLM Leaderboard data85.7%
22GPT-4.1 mini· OpenVLM Leaderboard data84.7%
23Gemma 3 12B IT· OpenVLM Leaderboard data83.9%
24Llama 3.2 11B Vision Instruct· OpenVLM Leaderboard data83.9%
25Gemini 1.5 Flash· OpenVLM Leaderboard data83.3%
26gemini-1.0-pro· OpenVLM Leaderboard data81.4%
27qwen2.5-vl-3b-instruct· OpenVLM Leaderboard data81.4%
28Qwen-VL Max· OpenVLM Leaderboard data80%
29GPT-4.1 nano· OpenVLM Leaderboard data77.5%
30Gemma 3 4B· OpenVLM Leaderboard data77.1%
31Qwen-VL Plus· OpenVLM Leaderboard data73.4%
32Molmo 7B-D· OpenVLM Leaderboard data58.1%
33kosmos-2· OpenVLM Leaderboard data32.7%
34Moonshot V1 8K· OpenVLM Leaderboard data1.2%

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