Vision & documents

MLVU (M-Avg) leaderboard

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

As of 2026-10-10, the highest measured score on MLVU (M-Avg) is 90.8% by Qwen3.8 Max. 64 more models have estimated scores, calibrated from the benchmarks they were measured on.

Measured scores: benchlm.ai.

#	Model	Score	Source
1	Qwen3.8 Max	90.8%	measured
2	GPT-6 Astra	90.4%	estimated ± 2.9 pp, low confidence
3	Claude Mythos 5	90.1%	estimated ± 1.5 pp, high confidence
4	Kimi K3	89.4%	estimated ± 1.5 pp, high confidence
5	Claude Opus 4.7 (Adaptive)	89.3%	estimated ± 1.5 pp, high confidence
6	Claude Opus 4.8	88.9%	estimated ± 1.5 pp, high confidence
7	GLM-5.3-Flash	88.8%	estimated ± 1.5 pp, high confidence
8	Qwen3.8-Flash-Next	88.7%	estimated ± 1.5 pp, medium confidence
9	Gemini 3.7 Flash	88.5%	estimated ± 1.5 pp, high confidence
10	Muse Spark 1.1	88.4%	estimated ± 1.5 pp, high confidence
11	Claude Sonnet 5	88.4%	estimated ± 1.5 pp, high confidence
12	Qwen3.8-Omni-Flash	88.2%	estimated ± 1.5 pp, high confidence
13	GPT-5.6 Sol	88.0%	estimated ± 2.7 pp, low confidence
14	Sakana Fugu-Ultra	87.8%	estimated ± 1.5 pp, high confidence
15	Muse Spark	87.8%	estimated ± 1.5 pp, high confidence
16	Seed 2.1 Pro	87.4%	estimated ± 1.5 pp, high confidence
17	Qwen3.7 Plus	87.4%	measured
18	Sakana Fugu	87.3%	estimated ± 1.5 pp, high confidence
19	GPT-5.5	87.3%	estimated ± 2.7 pp, medium confidence
20	Qwen3.8-27B	87.2%	estimated ± 1.5 pp, high confidence
21	GPT-5.6 Terra	87.1%	estimated ± 2.7 pp, medium confidence
22	Gemini 3.5 Flash	87.0%	estimated ± 1.5 pp, high confidence
23	Qwen3.5 397B	86.7%	measured
24	Qwen3.6-27B	86.6%	measured
25	GPT-5.4	86.6%	estimated ± 1.5 pp, high confidence
26	Seed 2.1 Turbo	86.5%	estimated ± 1.5 pp, high confidence
27	Inkling	86.3%	estimated ± 1.5 pp, high confidence
28	dots3-note Preview	86.2%	estimated ± 1.6 pp, medium confidence
29	Qwen3.6-35B-A3B	86.2%	measured
30	Qwen3.6 Plus	86.2%	estimated ± 1.5 pp, high confidence
31	GPT-5.6 Luna	86.1%	estimated ± 2.7 pp, medium confidence
32	Holo2-235B-A22B	86.1%	estimated ± 2.9 pp, medium confidence
33	Inkling-Small	86.1%	estimated ± 1.5 pp, high confidence
34	Step 3.7 Flash	86.1%	estimated ± 2.2 pp, medium confidence
35	Grok 4.3	86.0%	estimated ± 2.7 pp, medium confidence
36	MiMo-V2.5	86.0%	estimated ± 1.5 pp, high confidence
37	Pareto 26.9	86.0%	estimated ± 2.7 pp, medium confidence
38	Nemotron 3 Nano Omni 30B A3B	85.8%	estimated ± 0.4 pp, medium confidence
39	Kimi K2.6	85.8%	estimated ± 1.5 pp, high confidence
40	Gemini 3.1 Pro	85.7%	estimated ± 1.5 pp, high confidence
41	GPT-5.2	85.6%	measured
42	Gemma 4 31B	85.5%	estimated ± 2.7 pp, medium confidence
43	Qwen3.5-27B	85.5%	estimated ± 1.5 pp, high confidence
44	GPT-5.4 mini	85.4%	estimated ± 2.7 pp, medium confidence
45	MiniMax M3	85.3%	estimated ± 2.1 pp, high confidence
46	Muse Glimmer 30B	85.3%	estimated ± 1.5 pp, high confidence
47	Holo2-30B-A3B	85.3%	estimated ± 2.9 pp, medium confidence
48	Step 5 Preview	85.1%	estimated ± 2.7 pp, medium confidence
49	Kimi K2.5	85.0%	measured
50	Kimi K2.5 (Reasoning)	84.8%	estimated ± 1.5 pp, high confidence
51	Qwen3.5-35B-A3B	84.8%	estimated ± 1.5 pp, high confidence
52	Claude Sonnet 4.6	84.8%	estimated ± 1.5 pp, high confidence
53	LFM2.5-VL-3B	84.8%	estimated ± 0.4 pp, medium confidence
54	Qwen3.5-122B-A10B	84.7%	estimated ± 1.5 pp, high confidence
55	Gemma 4 26B A4B	84.2%	estimated ± 2.7 pp, medium confidence
56	Holo2-8B	83.9%	estimated ± 2.9 pp, medium confidence
57	Ternary Bonsai 2 27B	83.8%	estimated ± 1.5 pp, high confidence
58	Holo2-4B	83.5%	estimated ± 2.9 pp, medium confidence
59	Gemini 3.1 Flash-Lite	83.4%	estimated ± 1.5 pp, high confidence
60	ZAYA1-VL-8B	83.2%	estimated ± 0.4 pp, medium confidence
61	Gemma 4 12B	83.1%	estimated ± 1.6 pp, medium confidence
62	Gemini 3 Pro	83.0%	measured
63	Claude Opus 4.6	83.0%	estimated ± 2.5 pp, medium confidence
64	Interfaze Beta	82.3%	estimated ± 0.4 pp, low confidence
65	Claude Opus 4.5	81.7%	measured
66	GPT-5.4 nano	80.5%	estimated ± 2.7 pp, low confidence
67	Grok 4.20	79.4%	estimated ± 1.5 pp, medium confidence
68	Command A+	76.7%	estimated ± 1.5 pp, medium confidence
69	Gemma 4 E4B	72.6%	estimated ± 2.7 pp, low confidence
70	North Micro Vision Instruct	71.8%	estimated ± 1.5 pp, medium confidence
71	Qwen2.5-VL-32B	70.5%	estimated ± 2.7 pp, low confidence
72	LFM2.5-VL-450M	68.9%	estimated ± 1.5 pp, medium confidence
73	Gemma 4 E2B	66.6%	estimated ± 2.7 pp, low confidence

MLVU (M-Avg) leaderboard

All leaderboards

Agentic · terminal

Agentic · tools

Coding

Math

Knowledge & reasoning

Instruction following

Multilingual

Vision & documents

Long context

General