BenchLeader
Menu
Best LLMsChoose a modelLeaderboardsModelsBenchmarksProvidersReleases9

MMMU-Pro

College-level multimodal understanding. Run by Artificial Analysis.

As of 25 Oct 2026, Claude Opus 5.5 leads MMMU-Pro on BenchLeader with 87.7%, ahead of GPT-6 Astra at 86.9%, across 282 model configurations with a published result.

Published by
Artificial Analysis
Category
Multimodal
Index weight
1.0
Models
282
Data as of
25 Oct 2026
Newest result
23 Oct 2026

Source: Artificial Analysis (artificialanalysis.ai). Data taken from the public leaderboard.

What the test looks like

MMMU-Pro is a harder version of the college-level multimodal exam MMMU: questions with images from science, engineering, medicine and the arts, with ten answer options and a vision-only variant where the question itself is inside the image.

How it is scored

Percent correct; Artificial Analysis runs the standard setting itself.

What to keep in mind

Requires vision, so text-only models are absent. Ten options make guessing less effective than on the original MMMU, so scores are lower.

Top 15
  1. 1Claude Opus 5.5 (max)87.7%
  2. 2GPT-6 Astra (max)86.9%
  3. 3Claude Opus 5.5 (xhigh)86.6%
  4. 4GPT-6 Astra (high)86.4%
  5. 5GPT-6 Astra (xhigh)86.2%
  6. 6GPT-6.1 Sol (max)86.0%
  7. 7Claude Opus 5.5 (high)85.8%
  8. 8Claude Opus 5.5 (medium)85.7%
  9. 9Gemini 3.8 Flash (high)85.6%
  10. 10Gemini 3.7 Flash (high)85.5%
  11. 11GPT-6.1 Sol (xhigh)85.1%
  12. 12GPT-6 Astra (medium)85.1%
  13. 13GPT-6.1 Sol (high)84.9%
  14. 14Gemini 3.7 Flash (low)84.9%
  15. 15Claude Opus 5 (max)84.7%
282 of 282
#
1Claude Opus 5.5max87.7%72.62026-09-22
2GPT-6 Astramax86.9%70.42026-09-03
3Claude Opus 5.5xhigh86.6%70.02026-09-22
4GPT-6 Astrahigh86.4%69.62026-09-03
5GPT-6 Astraxhigh86.2%68.92026-09-03
6GPT-6.1 Solmax86.0%69.92026-09-29
7Claude Opus 5.5high85.8%69.42026-09-22
8Claude Opus 5.5medium85.7%66.52026-09-22
9Gemini 3.8 Flashhigh85.6%64.42026-09-02
10Gemini 3.7 Flashhigh85.5%63.52026-08-13
11GPT-6.1 Solxhigh85.1%67.72026-09-29
12GPT-6 Astramedium85.1%67.52026-09-03
13GPT-6.1 Solhigh84.9%67.72026-09-29
14Gemini 3.7 Flashlow84.9%60.32026-08-13
15Claude Opus 5max84.7%68.82026-07-24
16Gemini 3.7 Flashmedium84.7%63.12026-08-13
17Claude Opus 5.5low84.7%64.72026-09-22
18GPT-6 Astralow84.6%65.92026-09-03
19Gemini 3.8 Flashlow84.5%61.02026-09-02
20Gemini 3.5 Flashhigh84.3%63.12026-05-19
21Gemini 3.8 Flashmedium84.2%64.42026-09-02
22Claude Opus 5xhigh84.0%67.42026-07-24
23GPT-6.1 Solmedium83.9%67.22026-09-29
24NewKimi K3.1 Pro83.9%66.92026-10-23
25Gemini 3.5 Flashmedium83.9%62.82026-05-19
26GPT-5.6 Solmax83.4%68.02026-07-09
27Gemini 3.6 Flashhigh83.2%62.02026-07-21
28GPT-6.1 Sollow83.1%64.92026-09-29
29GPT-6 Solmax83.0%66.42026-09-22
30Qwen3.8 Maxmax82.8%63.32026-09-02
31GPT-5.6 Solxhigh82.7%66.62026-07-09
32GPT-6 Solxhigh82.5%64.32026-09-22
33Claude Opus 5high82.4%67.92026-07-24
34Gemini 3.1 Pro82.4%60.52026-02-19
35Muse Spark 1.3xhigh82.0%64.92026-09-02
36GPT-6 Solhigh82.0%63.72026-09-22
37GPT-5.6 Solhigh81.8%66.32026-07-09
38Claude Opus 5medium81.6%64.62026-07-24
39GPT-5.6 Solmedium81.4%64.42026-07-09
40GPT-5.5medium81.2%62.92026-04-23
41GPT-5.5high81.1%65.32026-04-23
42GPT-5.6 Sollow81.0%62.22026-07-09
43Muse Spark Openxhigh80.9%62.02026-10-15
44GPT-5.6 Terramax80.7%64.42026-07-09
45DeepSeek V4.1 Flash 1017max80.6%64.02026-10-17
46Kimi K3max80.5%64.82026-07-16
47Muse Spark80.5%64.32026-04-08
48Qwen3.7 Plus80.5%58.52026-06-01
49GPT-6 Solmedium80.4%62.62026-09-22
50Grok 4.5high80.4%59.32026-07-08
51Gemini 3 Prohigh80.2%60.02025-11-18
52GPT-6 Sollow80.2%59.72026-09-22
53Gemini 3.5 Flashminimal80.1%56.42026-05-19
54Gemini 3 Flashthinking79.9%59.12025-12-17
55GPT-5.5xhigh79.9%66.12026-04-23
56Claude Opus 5low79.8%61.32026-07-24
57Qwen3.8-Flash-Next79.8%60.32026-08-26
58GPT-6 Lunaxhigh79.7%58.72026-09-22
59GPT-6 Lunamax79.7%60.02026-09-22
60GPT-5.6 Terraxhigh79.5%62.82026-07-09

More multimodal benchmarks · All LLM benchmarks

Cite as: BenchLeader, “MMMU-Pro leaderboard”, data as of 25 Oct 2026, https://www.benchleader.com/benchmarks/aa_mmmu_pro.

MMMU-Pro: questions

What does MMMU-Pro measure?
MMMU-Pro is a harder version of the college-level multimodal exam MMMU: questions with images from science, engineering, medicine and the arts, with ten answer options and a vision-only variant where the question itself is inside the image. Scores are reported in percent of tasks solved; higher is better.
Which AI model leads MMMU-Pro?
Claude Opus 5.5 leads MMMU-Pro with 87.7% as of 25 Oct 2026, ahead of GPT-6 Astra at 86.9%.
How many models have MMMU-Pro results?
282 model configurations have a MMMU-Pro result on BenchLeader, all taken from Artificial Analysis.
Who runs MMMU-Pro and how often is it updated?
MMMU-Pro is published by Artificial Analysis. BenchLeader re-reads the published results once a day and records the date each result was published.
Does MMMU-Pro count toward the BenchLeader Index?
Yes. MMMU-Pro contributes to the multimodal category of the BenchLeader Index, normalised so that 50 is the average of the evaluated models.