CharXiv Reasoning

A scientific chart reasoning benchmark that tests whether models can understand, interpret, and reason about complex scientific visualizations including plots, diagrams, and data charts.

In index
Multimodal

Models39

Top result93.5%Claude Mythos 5

Top-3 spread2.1 pts

YearN/A

Result and price

Results

39 results

#ModelResult

01

Claude Mythos 593.5%

01

Qwen3.8 Max93.5%

03

Qwen3.8-Omni-Flash91.4%

04MAKimi K391.3%

05

Claude Opus 4.7 (Adaptive)91.0%

06

Qwen3.8-Flash-Next90.6%

07

Qwen3.8-27B90.2%

08

Claude Opus 4.889.9%

09

GLM-5.3-Flash89.4%

10

Gemini 3.7 Flash88.7%

11

Muse Spark 1.188.4%

12

Claude Sonnet 588.3%

13SASakana Fugu-Ultra86.6%

14

Muse Spark86.4%

15

Qwen3.7 Plus85.9%

16

Seed 2.1 Pro85.4%

17SASakana Fugu85.1%

18

Gemini 3.5 Flash84.2%

19

GPT-5.482.8%

20

Seed 2.1 Turbo82.5%

21

GPT-5.282.1%

22TMInkling82.0%

23

Qwen3.6 Plus81.5%

24

Gemini 3 Pro81.4%

25TMInkling-Small81.3%

26

MiMo-V2.581.0%

27

Qwen3.5 397B80.8%

28MAKimi K2.680.4%

29

Gemini 3.1 Pro80.2%

30

Muse Glimmer 30B78.8%

31

Qwen3.6-27B78.4%

32

Qwen3.6-35B-A3B78.0%

33

Claude Sonnet 4.677.4%

34

Qwen3.5-122B-A10B77.2%

35

Nemotron 3 Nano Omni 30B A3B76.3%

36

Gemini 3.1 Flash-Lite73.2%

37

Claude Opus 4.568.5%

38

Grok 4.2060.9%

39COCommand A+52.7%