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
Best in multimodal

Models49

Top result93.5%Claude Mythos 5

Top-3 spread0.1 pts

Year—

Result and price

40 of 49 results

Results

49 results

#ModelResult

01Claude Mythos 593.5%

01Qwen3.8 Max93.5%

03Kimi K3.1 Pro93.4%

04Qwen3.8-Omni-Flash91.4%

05Kimi K391.3%

06Claude Opus 4.7 (Adaptive)91.0%

07Qwen3.8-Flash-Next90.6%

08Qwen3.8-27B90.2%

08LongCat-2.590.2%

10Claude Opus 4.889.9%

11GLM-5.3-Flash89.4%

12Gemini 3.7 Flash88.7%

13Muse Spark 1.188.4%

14Claude Sonnet 588.3%

15Inkling 2 Small87.7%

16Sakana Fugu-Ultra86.6%

17Muse Spark86.4%

18Qwen3.7 Plus85.9%

19Seed 2.1 Pro85.4%

20Sakana Fugu85.1%

21Gemini 3.5 Flash84.2%

22GPT-5.482.8%

23MiniMax M3.1 Flash82.5%

23Seed 2.1 Turbo82.5%

25GPT-5.282.1%

26Inkling82.0%

27Qwen3.6 Plus81.5%

28Gemini 3 Pro81.4%

29Inkling-Small81.3%

30MiMo-V2.581.0%

31Nemotron 3.5 Super VL 120B A12B80.8%

31Qwen3.5 397B80.8%

33Kimi K2.680.4%

34Gemini 3.1 Pro80.2%

35Nemotron 3.5 Super VL 120B A12B NVFP479.9%

36Qwen3.5-27B79.5%

37Muse Glimmer 30B78.8%

38Qwen3.6-27B78.4%

39Qwen3.6-35B-A3B78.0%

40Kimi K2.5 (Reasoning)77.5%

40Kimi K2.577.5%

40Qwen3.5-35B-A3B77.5%

43Claude Sonnet 4.677.4%

44Qwen3.5-122B-A10B77.2%

45Nemotron 3 Nano Omni 30B A3B76.3%

46Gemini 3.1 Flash-Lite73.2%

47Claude Opus 4.568.5%

48Grok 4.2060.9%

49Command A+52.7%