benchgap
Vision & documents

MathVision leaderboard

As of 2026-11-21, the highest measured score on MathVision is 95.2% by Qwen3.8 Max. 115 more models have estimated scores, calibrated from the benchmarks they were measured on.

Measured scores: benchlm.ai.

#ModelScoreSource
1Gemini 3.1 Pro100.0%estimated ± 2.2 pp, low confidence
2Muse Spark100.0%estimated ± 2.2 pp, low confidence
3GPT-6.2 Astra96.9%estimated ± 3.1 pp, medium confidence
4Claude Opus 5.596.0%estimated ± 3.1 pp, medium confidence
5GPT-6.1 Terra95.6%estimated ± 3.0 pp, medium confidence
6GPT-6 Astra95.5%estimated ± 3.1 pp, medium confidence
7Kimi K3.1 Pro95.3%estimated ± 2.7 pp, low confidence
8Qwen3.8 Max95.2%measured
9GPT-6.1 Sol95.0%estimated ± 3.1 pp, medium confidence
10Gemini 3.8 Flash94.8%estimated ± 3.1 pp, medium confidence
11Gemini 3.7 Flash94.7%estimated ± 3.1 pp, medium confidence
12Kimi K394.3%measured
13Claude Opus 594.2%estimated ± 3.1 pp, medium confidence
14Muse Spark Open93.9%estimated ± 0.9 pp, high confidence
15Gemini 3.5 Flash93.5%estimated ± 3.0 pp, medium confidence
16Muse Spark 1.193.3%estimated ± 0.9 pp, high confidence
17Gemini 3.6 Flash93.2%estimated ± 3.1 pp, medium confidence
18Claude Mythos 593.1%estimated ± 3.3 pp, high confidence
19GPT-6 Sol92.9%estimated ± 3.1 pp, medium confidence
20Qwen3.8 Max Preview92.9%estimated ± 3.1 pp, medium confidence
21GPT-5.6 Sol92.9%estimated ± 3.0 pp, medium confidence
22Inkling 2 Small92.8%measured
23Glimpse92.8%estimated ± 3.3 pp, high confidence
24Seed 2.1 Pro92.6%measured
25LongCat-2.592.5%estimated ± 0.7 pp, medium confidence
26Qwen3.8-Omni-Flash91.8%measured
27Claude Opus 4.891.2%estimated ± 3.3 pp, high confidence
28Grok 4.590.9%estimated ± 3.1 pp, high confidence
29GPT-5.490.9%estimated ± 3.0 pp, high confidence
30GPT-5.590.9%estimated ± 3.0 pp, high confidence
31Qwen3.8-Flash-Next90.6%measured
32GPT-5.6 Terra90.3%estimated ± 3.0 pp, high confidence
33GPT-6 Luna90.3%estimated ± 3.1 pp, high confidence
34Qwen3.7 Plus90.3%measured
35Seed 2.1 Turbo90.1%measured
36Qwen3.8-27B90.0%measured
37Apodex 1.189.8%estimated ± 3.1 pp, high confidence
38Apodex 1.1 Mini89.8%estimated ± 3.1 pp, high confidence
39MiniMax M3.1 Flash89.7%estimated ± 3.1 pp, high confidence
40Gemini 3.5 Flash-Lite89.6%estimated ± 3.1 pp, high confidence
41Ling 3.0 Flash VL89.6%estimated ± 3.1 pp, high confidence
42Claude Opus 4.789.5%estimated ± 3.1 pp, high confidence
43Sakana Fugu-Ultra89.3%estimated ± 3.3 pp, high confidence
44Gemini 3 Flash89.3%estimated ± 3.1 pp, high confidence
45GPT-5.3 Codex89.2%estimated ± 3.1 pp, high confidence
46Inkling 2 Mini89.0%measured
47Qwen3.5 397B88.6%measured
48GLM-5.3-Flash88.6%estimated ± 0.9 pp, high confidence
49Sakana Fugu88.4%estimated ± 3.3 pp, high confidence
50Qwen3.6 Plus88.0%measured
51Hy488.0%estimated ± 3.0 pp, high confidence
52Claude Sonnet 587.9%estimated ± 3.1 pp, high confidence
53GPT-5.6 Luna87.8%estimated ± 3.0 pp, high confidence
54dots3-note Preview87.7%measured
55DeepSeek V4.1 Flash 101787.6%estimated ± 3.1 pp, high confidence
56Grok 4.387.4%estimated ± 3.0 pp, high confidence
57MiniMax M387.4%estimated ± 3.0 pp, high confidence
58Kimi K2.687.4%measured
59Pareto 26.987.3%estimated ± 3.0 pp, high confidence
60MiMo-V2.587.2%estimated ± 3.0 pp, high confidence
61GPT-5.2-Codex86.8%estimated ± 3.1 pp, high confidence
62Gemini 3 Pro86.6%measured
63Claude Opus 4.686.5%estimated ± 3.0 pp, high confidence
64Qwen3.5-122B-A10B86.2%measured
65Gemma 4 31B86.1%estimated ± 3.0 pp, high confidence
66Qwen3.5-27B86.0%measured
67GPT-5.185.8%estimated ± 3.1 pp, high confidence
68GPT-5.4 mini85.8%estimated ± 3.0 pp, high confidence
69Claude Opus 4.6 (Adaptive)85.7%estimated ± 3.1 pp, high confidence
70Step 3.7 Flash85.6%estimated ± 3.1 pp, high confidence
71Qwen3.6-35B-A3B85.2%estimated ± 2.6 pp, medium confidence
72Gemini 2.5 Pro85.1%estimated ± 3.1 pp, high confidence
73Step 5 Preview85.1%estimated ± 3.0 pp, high confidence
74Holo2-235B-A22B85.0%estimated ± 10.9 pp, low confidence
75Hy4 flash84.9%estimated ± 3.0 pp, high confidence
76Qwen3.6-27B84.9%estimated ± 3.0 pp, high confidence
77GPT-584.2%estimated ± 3.1 pp, high confidence
78Kimi K2.584.2%measured
79Grok 4.2084.2%estimated ± 3.0 pp, high confidence
80Claude Opus 4.5 Thinking83.9%estimated ± 3.1 pp, high confidence
81Qwen3.5-35B-A3B83.9%measured
82GPT-5.283.0%measured
83Inkling-Small82.9%estimated ± 3.0 pp, high confidence
84Muse Glimmer 30B82.9%estimated ± 3.0 pp, high confidence
85MiMo-V2.6-Flash82.7%estimated ± 3.1 pp, high confidence
86Gemma 4 26B A4B82.6%estimated ± 3.0 pp, high confidence
87Inkling82.3%estimated ± 3.0 pp, high confidence
88GLM-5V-Turbo82.3%estimated ± 3.1 pp, high confidence
89GPT-5.1-Codex81.8%estimated ± 3.1 pp, high confidence
90GPT-5.1-Codex-Max81.8%estimated ± 3.1 pp, high confidence
91Holo2-30B-A3B81.8%estimated ± 10.9 pp, low confidence
92Gemini 3.1 Flash-Lite81.1%estimated ± 3.3 pp, high confidence
93Gemma 4 12B79.7%measured
94Interfaze Beta79.6%estimated ± 3.0 pp, high confidence
95Claude Sonnet 4.678.9%estimated ± 3.1 pp, high confidence
96Nemotron 3.5 Super VL 120B A12B78.2%measured
97o378.1%estimated ± 3.1 pp, high confidence
98Nemotron 3.5 Super VL 120B A12B NVFP477.8%measured
99MiMo-V2-Omni77.7%estimated ± 3.1 pp, high confidence
100Nemotron 3 Nano Omni 30B A3B76.9%estimated ± 2.6 pp, medium confidence
101Ternary Bonsai 2 27B76.8%estimated ± 9.1 pp, medium confidence
102o4-mini76.5%estimated ± 3.1 pp, medium confidence
103Qwen3-VL-235B-A22B Instruct76.3%estimated ± 3.0 pp, high confidence
104Holo2-8B76.3%estimated ± 10.9 pp, low confidence
105Grok 475.8%estimated ± 3.1 pp, medium confidence
106Claude 4 Opus Thinking75.7%estimated ± 3.6 pp, high confidence
107ZAYA1-VL-8B75.1%estimated ± 2.6 pp, low confidence
108Holo2-4B74.9%estimated ± 10.9 pp, low confidence
109Claude Opus 4.574.3%measured
110GPT-5.4 nano74.1%estimated ± 3.0 pp, high confidence
111GLM-4.6V74.0%estimated ± 3.0 pp, high confidence
112Claude 3.7 Sonnet Thinking74.0%estimated ± 3.6 pp, high confidence
113Claude 4 Opus72.5%estimated ± 3.6 pp, high confidence
114Gemini 2.5 Flash-Lite71.5%estimated ± 3.6 pp, high confidence
115Command A+70.7%estimated ± 3.0 pp, high confidence
116Gemini 2.5 Flash69.3%estimated ± 3.1 pp, medium confidence
117Qwen3.5-4B69.0%estimated ± 3.1 pp, medium confidence
118GLM-4.6V-Flash68.0%estimated ± 3.0 pp, high confidence
119Mistral Medium 3.5 128B68.0%estimated ± 3.1 pp, medium confidence
120Gemini 2.0 Flash67.4%estimated ± 3.6 pp, high confidence
121Grok 4.1 Fast (Reasoning)64.3%estimated ± 3.1 pp, medium confidence
122Grok 263.7%estimated ± 3.6 pp, high confidence
123Gemini 2.0 Flash-Lite62.5%estimated ± 3.6 pp, high confidence
124Claude 4 Sonnet62.2%estimated ± 3.1 pp, medium confidence
125Llama 4 Maverick61.5%estimated ± 3.1 pp, medium confidence
126Grok 4 Fast (Reasoning)60.7%estimated ± 3.1 pp, medium confidence
127Grok 2 mini60.4%estimated ± 3.6 pp, high confidence
128GPT-4.159.2%estimated ± 3.1 pp, medium confidence
129MiniMax-VL-0159.2%estimated ± 3.0 pp, high confidence
130Gemma 4 E4B59.1%estimated ± 3.0 pp, high confidence
131Qwen3-Omni-30B-A3B-Thinking56.7%estimated ± 3.1 pp, medium confidence
132Claude 3.7 Sonnet56.4%estimated ± 3.1 pp, medium confidence
133Qwen2.5-VL-32B55.7%estimated ± 3.0 pp, high confidence
134Mistral Small 3.155.4%estimated ± 3.0 pp, high confidence
135Gemini 1.5 Flash54.7%estimated ± 3.6 pp, high confidence
136GPT-4.1 mini52.7%estimated ± 3.1 pp, medium confidence
137LFM3-VL-2B51.6%measured
138Gemma 4 E2B49.8%estimated ± 3.0 pp, medium confidence
139LFM2.5-VL-3B34.6%estimated ± 3.0 pp, medium confidence
140LFM2.5-VL-450M25.2%estimated ± 3.6 pp, medium confidence

All leaderboards

Agentic · terminal

Agentic · tools

Coding

Math

Knowledge & reasoning

Instruction following

Multilingual

Vision & documents

Long context

General