Vision & documents

VideoMMMU leaderboard

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

As of 2026-10-07, the highest measured score on VideoMMMU is 88.7% by Qwen3.8 Max. 55 more models have estimated scores, calibrated from the benchmarks they were measured on.

Measured scores: benchlm.ai.

#	Model	Score	Source
1	GPT-6 Astra	90.6%	estimated ± 1.7 pp, low confidence
2	Gemini 3.1 Pro	89.3%	estimated ± 1.0 pp, medium confidence
3	Gemini 3.5 Flash	89.2%	estimated ± 1.0 pp, medium confidence
4	GPT-5.6 Sol	89.0%	estimated ± 1.0 pp, medium confidence
5	Qwen3.8 Max	88.7%	measured
6	Kimi K3	88.2%	estimated ± 1.0 pp, high confidence
7	Seed 2.1 Pro	88.2%	estimated ± 1.0 pp, high confidence
8	Claude Mythos 5	88.2%	estimated ± 1.5 pp, high confidence
9	GPT-5.4	87.8%	estimated ± 1.0 pp, high confidence
10	GPT-5.5	87.8%	estimated ± 1.0 pp, high confidence
11	Gemini 3 Pro	87.6%	measured
12	Qwen3.8-Omni-Flash	87.6%	estimated ± 1.5 pp, high confidence
13	Claude Opus 4.7 (Adaptive)	87.4%	estimated ± 1.5 pp, high confidence
14	Qwen3.8-Flash-Next	87.3%	estimated ± 1.5 pp, high confidence
15	GPT-5.6 Terra	87.3%	estimated ± 1.0 pp, high confidence
16	Qwen3.8-27B	87.2%	estimated ± 1.5 pp, high confidence
17	Claude Opus 4.8	87.1%	estimated ± 1.5 pp, high confidence
18	Step 3.7 Flash	87.0%	estimated ± 2.2 pp, low confidence
19	GLM-5.3-Flash	87.0%	estimated ± 1.5 pp, high confidence
20	Muse Spark	87.0%	estimated ± 1.0 pp, high confidence
21	Gemini 3.7 Flash	86.8%	estimated ± 1.5 pp, high confidence
22	dots3-note Preview	86.8%	measured
23	Muse Spark 1.1	86.8%	estimated ± 1.5 pp, high confidence
24	Claude Sonnet 5	86.7%	estimated ± 1.5 pp, high confidence
25	Seed 2.1 Turbo	86.6%	estimated ± 1.0 pp, high confidence
26	Kimi K2.5	86.6%	measured
27	Sakana Fugu-Ultra	86.3%	estimated ± 1.5 pp, high confidence
28	GPT-5.2	86.0%	estimated ± 1.0 pp, high confidence
29	Sakana Fugu	86.0%	estimated ± 1.5 pp, high confidence
30	Kimi K2.6	85.9%	estimated ± 1.0 pp, high confidence
31	Qwen3.5-27B	85.6%	estimated ± 1.9 pp, low confidence
32	Holo2-235B-A22B	85.5%	estimated ± 1.7 pp, medium confidence
33	Qwen3.7 Plus	85.4%	measured
34	Qwen3.5-35B-A3B	85.2%	estimated ± 1.9 pp, low confidence
35	Kimi K2.5 (Reasoning)	85.1%	estimated ± 1.0 pp, high confidence
36	GPT-5.6 Luna	85.1%	estimated ± 1.0 pp, high confidence
37	Holo2-30B-A3B	85.0%	estimated ± 1.7 pp, medium confidence
38	Grok 4.3	84.9%	estimated ± 1.0 pp, high confidence
39	Pareto 26.9	84.8%	estimated ± 1.0 pp, high confidence
40	MiMo-V2.5	84.8%	estimated ± 1.0 pp, high confidence
41	Qwen3.5 397B	84.7%	measured
42	MiniMax M3	84.6%	measured
43	Claude Sonnet 4.6	84.6%	estimated ± 1.5 pp, high confidence
44	Qwen3.5-122B-A10B	84.5%	estimated ± 1.5 pp, high confidence
45	Claude Opus 4.6	84.5%	estimated ± 1.0 pp, high confidence
46	Holo2-8B	84.4%	estimated ± 1.7 pp, medium confidence
47	Gemma 4 31B	84.4%	estimated ± 1.0 pp, high confidence
48	Nemotron 3 Nano Omni 30B A3B	84.4%	estimated ± 1.5 pp, high confidence
49	Claude Opus 4.5	84.4%	measured
50	Qwen3.6-27B	84.4%	measured
51	GPT-5.4 mini	84.4%	estimated ± 1.0 pp, high confidence
52	Holo2-4B	84.4%	estimated ± 1.7 pp, medium confidence
53	Step 5 Preview	84.3%	estimated ± 1.0 pp, high confidence
54	Grok 4.20	84.2%	estimated ± 1.0 pp, high confidence
55	Inkling-Small	84.1%	estimated ± 1.0 pp, high confidence
56	Muse Glimmer 30B	84.1%	estimated ± 1.0 pp, high confidence
57	Gemma 4 26B A4B	84.1%	estimated ± 1.0 pp, high confidence
58	Inkling	84.1%	estimated ± 1.0 pp, high confidence
59	Interfaze Beta	84.1%	estimated ± 1.0 pp, high confidence
60	Gemma 4 12B	84.1%	estimated ± 1.0 pp, medium confidence
61	Command A+	84.1%	estimated ± 1.0 pp, medium confidence
62	GPT-5.4 nano	84.1%	estimated ± 1.0 pp, medium confidence
63	LFM2.5-VL-3B	84.1%	estimated ± 1.0 pp, medium confidence
64	Gemini 3.1 Flash-Lite	84.0%	estimated ± 1.5 pp, high confidence
65	Qwen3.6 Plus	84.0%	measured
66	Qwen3.6-35B-A3B	83.7%	measured

VideoMMMU leaderboard

All leaderboards

Agentic · terminal

Agentic · tools

Coding

Math

Knowledge & reasoning

Instruction following

Multilingual

Vision & documents

Long context

General