LLM Reference
MMMUactive

MMMU: Massive Multi-discipline Multimodal Understanding

Metric: Accuracy (higher is better)Introduced: 2023

11,500+ vision-language questions spanning 30 disciplines across six core areas (art, business, science, health, humanities, tech). Evaluates college-level multimodal reasoning. High benchmark score alone doesn't make a model the right pick — weigh it against pricing, API availability, and release date.

Models ranked

46

tracked on this benchmark

Score band

86.0 – 30.3

best → lowest tracked

Snapshot trend

-12.30

May 12 → Jun 7 · 6 models

Leaderboard

Tracked models ranked by Accuracy (higher is better).

Compare candidates
#Model variant and provenanceScore
1
Qwen3.6-Plus
Version: MMMUHarness: Not recordedEvaluator: Not recordedObserved: May 12, 2026Confidence: Not recordedSource

Notes: Confidence: high. DAT-4172 May 12 /best/ refresh; llm-stats listed Qwen3.6 Plus at 86.0%.

86.0
2
ByteDance Doubao Seed 2.0 Pro
Version: MMMU (accuracy)Harness: Not recordedEvaluator: Not recordedObserved: Jun 7, 2026Confidence: Not recordedSource
85.4
3
Qwen3.5-397B-A17B
Version: Not recordedHarness: Not recordedEvaluator: Not recordedObserved: Apr 19, 2026Confidence: Not recordedSource
85.0
4
Gemini 3.5 Flash
Version: Not recordedHarness: Not recordedEvaluator: Not recordedObserved: May 19, 2026Confidence: Not recordedSource
83.6
5
Claude Sonnet 4.6
Version: MMMU (accuracy)Harness: Not recordedEvaluator: Not recordedObserved: Jun 7, 2026Confidence: Not recordedSource
83.6
6
o3
Version: standard MMMUHarness: Not recordedEvaluator: Not recordedObserved: Apr 16, 2025Confidence: Not recordedSource

Notes: DAT-5669: official o3 launch figure. Do not promote MMMU-Pro for this pair without official corroboration.

82.9
7
GPT-5.4
Version: Not recordedHarness: Not recordedEvaluator: Not recordedObserved: Apr 19, 2026Confidence: Not recordedSource
82.1
8
Qwen3.6 Max Preview
Version: MMMUHarness: Not recordedEvaluator: Not recordedObserved: May 12, 2026Confidence: Not recordedSource

Notes: Confidence: medium. DAT-4172 May 12 /best/ refresh; queued for recheck in the next audit.

82.0
9
Gemini 2.5 Pro
Version: standard MMMU pass@1, no extended thinkingHarness: Not recordedEvaluator: Not recordedObserved: Mar 25, 2025Confidence: Not recordedSource

Notes: DAT-5669: official standard MMMU figure. Gemini 2.5 Pro Deep Think 84.0% is a separate variant and should not replace this comparable row.

81.7
10
o4-mini
Version: Not recordedHarness: Not recordedEvaluator: Not recordedObserved: Apr 15, 2026Confidence: Not recordedSource
81.6
11
Gemini 3 Pro
Version: Not recordedHarness: Not recordedEvaluator: Not recordedObserved: Apr 19, 2026Confidence: Not recordedSource
81.0
12
Claude Opus 4.5
Version: Not recordedHarness: Not recordedEvaluator: Not recordedObserved: Apr 15, 2026Confidence: Not recordedSource
80.7
13
Gemini 2.5 Flash
Version: Not recordedHarness: Not recordedEvaluator: Not recordedObserved: Apr 15, 2026Confidence: Not recordedSource
79.7
14
Gemini 2.5 Pro Preview 05-06
Version: Not recordedHarness: Not recordedEvaluator: Not recordedObserved: Apr 15, 2026Confidence: Not recordedSource
79.6
15
o1 (12-17)
Version: Not recordedHarness: Not recordedEvaluator: Not recordedObserved: Apr 15, 2026Confidence: Not recordedSource
78.2
16
Grok-3
Version: Not recordedHarness: Not recordedEvaluator: Not recordedObserved: Apr 15, 2026Confidence: Not recordedSource
78.0
17
Claude Sonnet 4.5
Version: Not recordedHarness: Not recordedEvaluator: Not recordedObserved: Apr 15, 2026Confidence: Not recordedSource
77.8
18
Claude Opus 4.1
Version: Not recordedHarness: Not recordedEvaluator: Not recordedObserved: Apr 15, 2026Confidence: Not recordedSource
77.1
19
Claude Opus 4.6
Version: MMMU (accuracy)Harness: Not recordedEvaluator: Not recordedObserved: Jun 7, 2026Confidence: Not recordedSource
76.5
20
Command A+
Version: MMMU (accuracy)Harness: Not recordedEvaluator: Not recordedObserved: Jun 7, 2026Confidence: Not recordedSource
75.1
21
Claude 3.7 Sonnet
Version: Not recordedHarness: Not recordedEvaluator: Not recordedObserved: Apr 15, 2026Confidence: Not recordedSource
75.0
22
Claude Sonnet 4
Version: Not recordedHarness: Not recordedEvaluator: Not recordedObserved: Apr 15, 2026Confidence: Not recordedSource
74.4
23
Llama 4 Maverick 17B Instruct FP8
Version: Not recordedHarness: Not recordedEvaluator: Not recordedObserved: Apr 5, 2025Confidence: Not recordedSource
73.4
24
Gemini 2.0 Flash Experimental
Version: Not recordedHarness: Not recordedEvaluator: Not recordedObserved: Apr 15, 2026Confidence: Not recordedSource
71.7
25
GPT-4o (2024-11-20)
Version: Not recordedHarness: Not recordedEvaluator: Not recordedObserved: Apr 15, 2026Confidence: Not recordedSource
70.7
Showing the top 25 of 46 tracked models. Browse all models.

How to read this benchmark

This benchmark scores models where higher is better. Scores are useful for directional filtering and shortlisting — not for universal quality ranking. Prefer benchmarks closest to your workload, then validate the linked model pages for pricing, context window, and provider availability.

Trust this score when

  • There is a fresh timestamped snapshot (or multiple snapshots) for this benchmark.
  • The model list covers the same version family you can actually deploy today.
  • Top candidates overlap with your required routing and feature requirements.

Be cautious when

  • There is only one benchmark snapshot or the dataset appears stale.
  • The benchmark metric direction is opposite of your decision objective.
  • The score difference between options is narrow and likely within implementation variance.

FAQ

What does the MMMU benchmark measure?

11,500+ vision-language questions spanning 30 disciplines across six core areas (art, business, science, health, humanities, tech). Evaluates college-level multimodal reasoning. On this page it lists 46 tracked model variants where higher is better.

Is a higher Massive Multi-discipline Multimodal Understanding score always better?

For this benchmark, higher is better. A high score helps you shortlist, but confirm pricing, context window, and provider availability on each model page before committing — the top scorer is not always the right pick for your workload or budget.

How current is this Massive Multi-discipline Multimodal Understanding data?

This benchmark was last reviewed on Apr 15, 2026. The tracked score average moved -12.30 points across the last 3 snapshots.

Last reviewed: Apr 15, 2026