Skip to content

Benchmark

xbench-ScienceQA

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

Models measured
30
Scores recorded
30
Sources
1
IQ weight
2.19

Score distribution

70.6%
Mean
71%
Median
3.1%
Std dev
65.8%
10th percentile
73.4%
90th percentile
64.6%
Min
64.6%76.4%
The band covers the middle 80% of scores. The line marks the median.

This benchmark has a weight of 2.19 for chat models, measured over 30 models. The weight is how much it counts toward llminfo IQ. We measure it from the data, not from the raw scores.

Leaderboard

#ModelScore
1Claude Opus 4.8default· xbench76.4%
2Claude Opus 5.5default· xbench76.4%
3GPT-6 Astramax, pro· xbench74.6%
4Claude Fable 5default· xbench73.4%
5Qwen3.8 Maxdefault· xbench73.4%
6Claude Opus 4.7default· xbench73%
7GPT-5.5 Prohigh· xbench73%
8DeepSeek Flashmax· xbench72.6%
9Gemini 3.8 Flashhigh· xbench72.4%
10Grok 4.5default· xbench72.4%
11GPT-5.4 Prodefault· xbench72.2%
12Gemini 3 Pro Previewdefault· xbench71.6%
13Kimi K3default· xbench71.2%
14DeepSeek V4 Flashdefault· xbench71%
15Doubao Seed 2.1 Prodefault· xbench71%
16GPT-5.6 Soldefault, pro· xbench71%
17GPT 5.4default· xbench70.8%
18Qwen3.7 Maxxhigh· xbench70.8%
19Grok 4.6default· xbench70.4%
20Gemini 3.7 Flashdefault· xbench70.2%
21Hy4 previewdefault· xbench69.8%
22DeepSeek V4 Promax· xbench69.4%
23Doubao Seed 2.0 Prohigh· xbench69.2%
24Qwen3.8 Max Previewdefault· xbench68%
25Hy3 (Deep Infra)default· xbench66.6%
26GLM 5.3default· xbench65.8%
27Kimi K2.6default· xbench65.8%
28Gemini 3.6 Flashdefault· xbench65.6%
29Grok 4default· xbench65%
30GLM 5.2default· xbench64.6%