live
weekly refresh
basedagi.org
benchmark evidence

MMMU

Massive Multidiscipline Multimodal Understanding — 11K expert questions across 57 subjects.

winner on MMMU
direct benchmark result, not a broad vertical composite | source row dated 2026-05-15
scored on 2026-05-15 · stale source data (55d)
latest mapped results | top 20
#ModelScoreEvidenceTested
1OpenAI: GPT-5.1
Openai
85.4
model-only
independent_benchmark
2026-05-15
2OpenAI: o3
Openai
82.9
model-only
independent_benchmark
2000-01-01
3OpenAI: o4 Mini
Openai
81.6
model-only
independent_benchmark
2000-01-01
4Google: Gemini 2.5 Flash
Google
79.7
model-only
independent_benchmark
2000-01-01
5Google: Gemini 2.5 Pro
Google
79.6
model-only
independent_benchmark
2000-01-01
6Anthropic: Claude Sonnet 4.5
Anthropic
77.8
model-only
independent_benchmark
2026-05-15
7OpenAI: o1
Openai
77.6
model-only
independent_benchmark
2000-01-01
8Anthropic: Claude Opus 4.7
Anthropic
76.5
model-only
independent_benchmark
2026-05-08
9Anthropic: Claude Opus 4.5
Anthropic
76.5
model-only
independent_benchmark
2026-05-15
10OpenAI: GPT-4.1
Openai
74.8
model-only
independent_benchmark
2000-01-01
11Anthropic: Claude Sonnet 4
Anthropic
74.4
model-only
independent_benchmark
2026-05-15
12Anthropic: Claude Sonnet 4.6
Anthropic
74.4
model-only
independent_benchmark
2026-05-08
13Meta: Llama 4 Maverick
Meta Llama
73.4
model-only
independent_benchmark
2026-05-15
14Google: Gemini 2.0 Flash Lite
Google
71.7
model-only
independent_benchmark
2026-05-15
15Google: Gemini 2.0 Flash
Google
70.7
model-only
independent_benchmark
2000-01-01
16Meta: Llama 4 Scout
Meta Llama
69.4
model-only
independent_benchmark
2026-05-15
17Google: Gemma 3 27B
Google
64.9
model-only
independent_benchmark
2026-05-15
18Amazon: Nova Pro 1.0
Amazon
62.0
model-only
independent_benchmark
2026-05-15
19OpenAI: GPT-5
Openai
56.8
model-only
independent_benchmark
2026-05-15
what this result means

Massive Multidiscipline Multimodal Understanding — 11K expert questions across 57 subjects.

This benchmark contributes direct public evidence. Read its scope before generalizing the result.

A win here is a win on MMMU. Broad task pages require independent corroboration before naming a general winner.

source record
category: reasoning
metric: accuracy
matched models: 19
latest source date: 2026-05-15
direction: higher is better
inspect upstream source ->