Benchmark profile

Vals MMMU (MMMU)

Multimodal Multi-task Benchmark

Data verified July 18, 2026

How BenchLM shows MMMU

BenchLM mirrors the public Vals AI MMMU leaderboard captured from https://www.vals.ai/benchmarks/mmmu and updated by Vals on July 18, 2026. The snapshot preserves overall scores, uncertainty, latency, cost-per-test metadata, and task-level scores where Vals publishes them.

MMMU is display only on BenchLM. Vals proprietary or Vals-hosted aggregate views are useful context, but BenchLM does not use them as weighted ranking inputs or as a replacement for benchmark-native source records.

83 Vals rows1 task viewspublic datasetTasks: OverallDisplay only

MMMU on Vals AI Vals methodology Vals home

MMMU score on MMMU — July 18, 2026

BenchLM mirrors the published mmmu score view for MMMU. Claude Fable 5 leads the public snapshot at 89.31% , followed by GPT-5.6 Sol (88.84%) and Gemini 3.5 Flash (88.27%). BenchLM does not use these results to rank models overall.

Claude Fable 5

Anthropic

anthropic/claude-fable-5

89.31%

Overall —

GPT-5.6 Sol

OpenAI

openai/gpt-5.6-sol

88.84%

Overall —

Gemini 3.5 Flash

Google

google/gemini-3.5-flash

88.27%

Overall —

83 modelsExternal benchmark mirrorsCurrentDisplay onlyUpdated July 18, 2026

MMMU score table (83 models)

Score

Claude Fable 5Anthropic

89.31%

GPT-5.6 SolOpenAI

88.84%

Gemini 3.5 FlashGoogle

88.27%

GPT-5.5OpenAI

88.27%

Gemini 3.1 Pro PreviewGoogle

88.21%

Kimi K3Moonshot AI

88.15%

Gemini 3 Flash PreviewGoogle

87.63%

Gemini 3 Pro PreviewGoogle

87.51%

GPT-5.4OpenAI

87.51%

Muse SparkMeta

87.40%

GPT-5.2OpenAI

86.67%

Muse Spark 1 1Meta

86.59%

Claude Opus 4.8Anthropic

86.59%

GPT-5.6 TerraOpenAI

86.47%

Kimi K2.6Moonshot AI

86.30%

Claude Opus 4.7Anthropic

85.55%

GPT-5.6 LunaOpenAI

85.03%

Kimi K2.5 ThinkingMoonshot AI

84.33%

Qwen3.6 PlusAlibaba

84.16%

Claude Opus 4.6 ThinkingAnthropic

83.87%

Claude Sonnet 4.6Anthropic

83.58%

Grok 4.20 0309 ReasoningSpaceXAI

83.47%

GPT-5.1OpenAI

83.18%

Grok 4.3SpaceXAI

83.06%

Claude Sonnet 5Anthropic

83.01%

Claude Opus 4.5 20251101 ThinkingAnthropic

82.95%

Gemini 3.1 Flash Lite PreviewGoogle

82.49%

Qwen3.5 FlashAlibaba

81.91%

GPT-5OpenAI

81.50%

Gemini 2.5 Pro Exp 03 25Google

81.34%

MiniMax M3MiniMax

81.16%

Claude Opus 4.5Anthropic

81.10%

Gemini 2.5 Flash Preview 09 2025 ThinkingGoogle

80.75%

O3OpenAI

80.42%

Mimo V2.5Xiaomi

80.00%

O4 MiniOpenAI

79.67%

Gemini 2.5 Flash Preview 09 2025Google

79.48%

Claude Sonnet 4.5 20250929 ThinkingAnthropic

79.31%

GPT-5.4 MiniOpenAI

79.25%

GPT-5 MiniOpenAI

78.91%

InklingThinkingmachines

78.50%

Claude Opus 4.1 20250805 ThinkingAnthropic

77.51%

O1OpenAI

77.41%

Grok 4 0709SpaceXAI

76.27%

Gemini 2.5 Flash Lite Preview 09 2025 ThinkingGoogle

75.43%

Claude 3.7 Sonnet 20250219 ThinkingAnthropic

75.10%

Claude Sonnet 4 20250514 ThinkingAnthropic

74.93%

Claude Opus 4.1Anthropic

73.72%

GPT-5.4 NanoOpenAI

73.58%

Claude Opus 4Anthropic

73.31%

Grok 4 Fast ReasoningSpaceXAI

72.78%

Grok 4.1 Fast ReasoningSpaceXAI

72.66%

Gemini 2.5 Flash Lite Preview 09 2025Google

72.54%

Claude Sonnet 4Anthropic

72.39%

GPT-4.1OpenAI

72.39%

Gemini 2.5 Flash Preview 04 17 ThinkingGoogle

71.92%

Llama4 Maverick Instruct BasicFireworks AI

71.69%

Claude 3.7 SonnetAnthropic

71.52%

GPT-5 NanoOpenAI

70.94%

GPT-4.1 MiniOpenAI

70.54%

Gemini 2.0 Flash 001Google

69.79%

Claude 3.5 SonnetAnthropic

68.80%

Gemini 2.5 Flash Preview 04 17Google

68.00%

Mistral Large 2512Mistral AI

66.19%

Gemini 1.5 Pro 002Google

65.51%

Magistral Small 2509Mistral AI

65.20%

Magistral Medium 2509Mistral AI

64.57%

GPT-4oOpenAI

64.01%

Grok 4.1 Fast Non ReasoningSpaceXAI

63.70%

Grok 4 Fast Non ReasoningSpaceXAI

63.41%

Mistral Medium 2505Mistral AI

62.97%

GPT-4oOpenAI

62.16%

Grok 4.5SpaceXAI

61.79%

Mistral Small 2503Mistral AI

60.08%

Meta Llama Llama 4 Scout 17B 16E InstructTogether AI

58.75%

Grok 2 Vision 1212SpaceXAI

57.25%

Gemini 1.5 Flash 002Google

57.19%

GPT-4o MiniOpenAI

56.56%

GPT-4.1 NanoOpenAI

55.05%

Meta Llama Llama 3.2 90B Vision Instruct TurboTogether AI

48.06%

Claude Haiku 4.5 20251001 ThinkingAnthropic

46.07%

Meta Llama Llama 3.2 11B Vision Instruct TurboTogether AI

38.82%

Qwen3.5 Plus ThinkingAlibaba

22.77%

The published MMMU snapshot places Claude Fable 5 first at 89.31%. The third row is 1.04 points behind. The broader top-10 range is 1.91 points, so many of the published results sit in a relatively narrow band.

83 models have been evaluated on MMMU. The benchmark falls in the External benchmark mirrors category. We keep external benchmark mirrors separate from the weighted global scoring system, so these results remain source-specific evidence. MMMU is currently displayed for reference but excluded from the scoring formula, so it does not directly affect overall rankings.

About MMMU

Year

2026

Tasks

Multimodal academic task suite

Format

Accuracy score

Difficulty

Multimodal college-level reasoning

BenchLM mirrors the public Vals AI MMMU leaderboard as display-only external evidence. The captured snapshot preserves overall scores, task-level scores where Vals publishes them, uncertainty, latency, and cost-per-test metadata. It is excluded from BenchLM weighted rankings.

MMMU Public benchmark source

BenchLM freshness & provenance

Version

MMMU 2026

Refresh cadence

Quarterly

Staleness state

Current

Question availability

Public benchmark set

CurrentDisplay only

BenchLM uses freshness metadata to decide whether a benchmark should still be treated as a strong differentiator, a benchmark to watch, or a display-only reference. For the full scoring policy, see the BenchLM methodology page.

FAQ

What does MMMU measure?

Multimodal Multi-task Benchmark

Which model leads the published MMMU snapshot?

Claude Fable 5 currently leads the published MMMU snapshot with 89.31% mmmu score. BenchLM shows this benchmark for display only and does not use it in overall rankings.

How many models are evaluated on MMMU?

83 AI models are included in BenchLM's mirrored MMMU snapshot, based on the public leaderboard captured on July 18, 2026.

Last updated: July 18, 2026 · mirrored from the public benchmark leaderboard

Choose a model with this week’s evidence

Join 2,000+ readers for ranking moves, pricing changes, and the claims that still need proof.

One email each week. Unsubscribe anytime.