Vision & documents

MathVision leaderboard

Name: benchgap: LLM benchmark scores, measured and estimated
Creator: benchgap

As of 2026-10-07, the highest measured score on MathVision is 95.2% by Qwen3.8 Max. 107 more models have estimated scores, calibrated from the benchmarks they were measured on.

Measured scores: benchlm.ai.

#	Model	Score	Source
1	Claude Mythos 5	96.3%	estimated ± 2.1 pp, low confidence
2	GPT-6 Astra	96.2%	estimated ± 2.4 pp, low confidence
3	GPT-5.6 Sol	95.6%	estimated ± 3.6 pp, medium confidence
4	Qwen3.8 Max	95.2%	measured
5	Kimi K3	94.3%	measured
6	Gemini 3.8 Flash	93.5%	estimated ± 2.1 pp, medium confidence
7	Muse Spark 1.1	93.3%	estimated ± 0.5 pp, medium confidence
8	Seed 2.1 Pro	92.6%	measured
9	GPT-5.5	92.1%	estimated ± 3.6 pp, high confidence
10	Qwen3.8-Omni-Flash	91.8%	measured
11	Gemini 3.7 Flash	91.5%	estimated ± 2.1 pp, medium confidence
12	GPT-5.6 Terra	91.2%	estimated ± 3.6 pp, high confidence
13	Qwen3.8-Flash-Next	90.6%	measured
14	Gemini 3.1 Pro	90.6%	estimated ± 1.4 pp, medium confidence
15	Muse Glimmer 30B	90.4%	estimated ± 2.4 pp, medium confidence
16	Kimi K2.5	90.3%	estimated ± 3.2 pp, medium confidence
17	Qwen3.7 Plus	90.3%	measured
18	Sakana Fugu-Ultra	90.1%	estimated ± 2.6 pp, high confidence
19	Seed 2.1 Turbo	90.1%	measured
20	Qwen3.8-27B	90.0%	measured
21	Claude Opus 5	89.7%	estimated ± 4.9 pp, medium confidence
22	Claude Opus 5.5	89.7%	estimated ± 4.9 pp, medium confidence
23	Gemini 3.6 Flash	89.7%	estimated ± 4.9 pp, medium confidence
24	GPT-6.1 Sol	89.7%	estimated ± 4.9 pp, medium confidence
25	GPT-6 Sol	89.7%	estimated ± 4.9 pp, medium confidence
26	Qwen3.8 Max Preview	89.7%	estimated ± 4.9 pp, medium confidence
27	Grok 4.5	89.7%	estimated ± 4.9 pp, high confidence
28	GPT-6 Luna	89.7%	estimated ± 4.9 pp, high confidence
29	Apodex 1.1	89.7%	estimated ± 4.9 pp, high confidence
30	Apodex 1.1 Mini	89.7%	estimated ± 4.9 pp, high confidence
31	Gemini 3.5 Flash-Lite	89.7%	estimated ± 4.9 pp, high confidence
32	Ling 3.0 Flash VL	89.7%	estimated ± 4.9 pp, high confidence
33	Gemini 3 Flash	89.6%	estimated ± 4.9 pp, high confidence
34	GPT-5.3 Codex	89.6%	estimated ± 4.9 pp, high confidence
35	DeepSeek V4.1 Flash	89.4%	estimated ± 4.9 pp, high confidence
36	Sakana Fugu	89.3%	estimated ± 2.6 pp, high confidence
37	GPT-5.4	89.2%	estimated ± 1.4 pp, low confidence
38	Claude Opus 4.7	89.1%	estimated ± 4.9 pp, high confidence
39	Mistral Large 4	89.1%	estimated ± 4.9 pp, high confidence
40	GPT-5.2-Codex	89.0%	estimated ± 4.9 pp, high confidence
41	Muse Spark	88.9%	estimated ± 1.4 pp, low confidence
42	Gemini 3.5 Flash	88.7%	estimated ± 2.6 pp, high confidence
43	Qwen3.5 397B	88.6%	measured
44	Holo2-235B-A22B	88.4%	estimated ± 2.4 pp, medium confidence
45	Claude Opus 4.7 (Adaptive)	88.3%	estimated ± 2.1 pp, low confidence
46	Qwen3.6-27B	88.3%	estimated ± 1.4 pp, low confidence
47	GLM-5.3-Flash	88.1%	estimated ± 0.5 pp, medium confidence
48	Qwen3.6 Plus	88.0%	measured
49	dots3-note Preview	87.7%	measured
50	GPT-5.1	87.7%	estimated ± 4.9 pp, high confidence
51	Claude Opus 4.6 (Adaptive)	87.5%	estimated ± 4.9 pp, high confidence
52	Kimi K2.6	87.4%	measured
53	Step 3.7 Flash	87.2%	estimated ± 2.7 pp, high confidence
54	Kimi K2.5 (Reasoning)	87.1%	estimated ± 3.6 pp, high confidence
55	GPT-5.6 Luna	86.9%	estimated ± 3.6 pp, high confidence
56	MiMo-V2.5	86.8%	estimated ± 2.6 pp, high confidence
57	Gemini 3 Pro	86.6%	measured
58	Grok 4.20	86.6%	estimated ± 1.4 pp, low confidence
59	Holo2-30B-A3B	86.5%	estimated ± 2.4 pp, medium confidence
60	Grok 4.3	86.3%	estimated ± 3.6 pp, high confidence
61	MiniMax M3	86.3%	estimated ± 3.6 pp, high confidence
62	Claude Opus 4.6	86.2%	estimated ± 1.4 pp, low confidence
63	Qwen3.5-122B-A10B	86.2%	measured
64	Pareto 26.9	86.2%	estimated ± 3.6 pp, high confidence
65	Gemini 2.5 Pro	86.0%	estimated ± 4.9 pp, high confidence
66	Qwen3.5-27B	86.0%	measured
67	Claude Opus 4.8	86.0%	estimated ± 2.1 pp, low confidence
68	Qwen3.6-35B-A3B	85.0%	estimated ± 2.6 pp, high confidence
69	Claude Sonnet 4.6	84.6%	estimated ± 2.6 pp, high confidence
70	GPT-5 (medium)	84.2%	estimated ± 4.9 pp, high confidence
71	Gemma 4 31B	84.2%	estimated ± 3.6 pp, high confidence
72	GPT-5 (high)	84.0%	estimated ± 4.9 pp, high confidence
73	Qwen3.5-35B-A3B	83.9%	measured
74	GPT-5.4 mini	83.7%	estimated ± 3.6 pp, high confidence
75	Claude Opus 4.5 Thinking	83.5%	estimated ± 4.9 pp, high confidence
76	GPT-5.2	83.0%	measured
77	Holo2-8B	82.9%	estimated ± 2.4 pp, medium confidence
78	Step 5 Preview	82.6%	estimated ± 3.6 pp, high confidence
79	Nemotron 3 Nano Omni 30B A3B	82.3%	estimated ± 2.4 pp, medium confidence
80	MiMo-V2.6-Flash	82.1%	estimated ± 4.9 pp, high confidence
81	Inkling	82.0%	estimated ± 2.1 pp, low confidence
82	Holo2-4B	82.0%	estimated ± 2.4 pp, medium confidence
83	Gemini 3.1 Flash-Lite	81.9%	estimated ± 2.6 pp, high confidence
84	GLM-5V-Turbo	81.9%	estimated ± 4.9 pp, high confidence
85	GPT-5.1-Codex	81.7%	estimated ± 4.9 pp, high confidence
86	GPT-5.1-Codex-Max	81.7%	estimated ± 4.9 pp, high confidence
87	o3	81.5%	estimated ± 4.9 pp, high confidence
88	MiMo-V2-Omni	81.5%	estimated ± 4.9 pp, high confidence
89	Grok 4	81.5%	estimated ± 4.9 pp, high confidence
90	Claude 4.1 Opus Thinking	81.5%	estimated ± 4.9 pp, high confidence
91	Claude 3 Haiku	81.5%	estimated ± 4.9 pp, medium confidence
92	Claude 4 Sonnet	81.5%	estimated ± 4.9 pp, high confidence
93	Gemini 1.5 Pro	81.5%	estimated ± 4.9 pp, high confidence
94	Gemini 2.5 Flash	81.5%	estimated ± 4.9 pp, high confidence
95	Gemma 3 27B	81.5%	estimated ± 4.9 pp, medium confidence
96	Gemma 4 E2B	81.5%	estimated ± 4.9 pp, medium confidence
97	Gemma 4 E4B	81.5%	estimated ± 4.9 pp, medium confidence
98	GPT-4.1	81.5%	estimated ± 4.9 pp, high confidence
99	GPT-4.1 mini	81.5%	estimated ± 4.9 pp, high confidence
100	GPT-4.1 nano	81.5%	estimated ± 4.9 pp, medium confidence
101	GPT-4o mini	81.5%	estimated ± 4.9 pp, medium confidence
102	Grok 4.1 Fast	81.5%	estimated ± 4.9 pp, medium confidence
103	Grok 4.1 Fast (Reasoning)	81.5%	estimated ± 4.9 pp, high confidence
104	Grok 4 Fast (Reasoning)	81.5%	estimated ± 4.9 pp, high confidence
105	LFM2.5-VL-1.6B-Extract	81.5%	estimated ± 4.9 pp, medium confidence
106	Llama 4 Maverick	81.5%	estimated ± 4.9 pp, high confidence
107	Llama 4 Scout	81.5%	estimated ± 4.9 pp, high confidence
108	Mistral Large 3	81.5%	estimated ± 4.9 pp, high confidence
109	Mistral Medium 3	81.5%	estimated ± 4.9 pp, high confidence
110	Mistral Medium 3.5 128B	81.5%	estimated ± 4.9 pp, high confidence
111	Mistral Small 4	81.5%	estimated ± 4.9 pp, high confidence
112	Mistral Small 4 (Reasoning)	81.5%	estimated ± 4.9 pp, high confidence
113	Nova Pro	81.5%	estimated ± 4.9 pp, medium confidence
114	Phi-4 Multimodal Instruct	81.5%	estimated ± 4.9 pp, medium confidence
115	Qwen3.5 397B (Reasoning)	81.5%	estimated ± 4.9 pp, high confidence
116	Qwen3-Omni-30B-A3B-Instruct	81.5%	estimated ± 4.9 pp, high confidence
117	Qwen3-Omni-30B-A3B-Thinking	81.5%	estimated ± 4.9 pp, high confidence
118	Inkling-Small	80.8%	estimated ± 2.1 pp, low confidence
119	Claude Sonnet 5	80.1%	estimated ± 2.1 pp, low confidence
120	Gemma 4 12B	79.7%	measured
121	Gemma 4 26B A4B	78.8%	estimated ± 3.6 pp, high confidence
122	Interfaze Beta	74.3%	estimated ± 3.6 pp, high confidence
123	Claude Opus 4.5	74.3%	measured
124	Command A+	66.6%	estimated ± 2.6 pp, medium confidence
125	GPT-5.4 nano	66.5%	estimated ± 3.6 pp, medium confidence
126	LFM2.5-VL-3B	24.3%	estimated ± 3.6 pp, medium confidence

MathVision leaderboard

All leaderboards

Agentic · terminal

Agentic · tools

Coding

Math

Knowledge & reasoning

Instruction following

Multilingual

Vision & documents

Long context

General