Artificial Analysis MMMU-Pro (AA-MMMU-Pro)

Data verified September 27, 202634 confirmed releases in the last 30 daysFollow model changes

An Artificial Analysis MMMU-Pro score.

Top models on AA-MMMU-Pro — September 27, 2026

As of September 27, 2026, Claude Opus 5.5 leads the AA-MMMU-Pro leaderboard with 87.7% , followed by GPT-6 Astra (86.9%) and Gemini 3.8 Flash (85.6%).

1Proprietary

Claude Opus 5.5

Anthropic

87.7%

Overall 86.98Context 1M

2Proprietary

GPT-6 Astra

OpenAI

86.9%

Overall 88.69Context 1.05M

3Proprietary

Gemini 3.8 Flash

Google

85.6%

Overall 73.43Context 1M

110 modelsMultimodal & Grounded15% of category scoreCurrentUpdated September 27, 2026

Leaderboard (110 models)

Score

Claude Opus 5.5Anthropic · Closed

87.7%

GPT-6 AstraOpenAI · Closed

86.9%

Gemini 3.8 FlashGoogle · Closed

85.6%

Gemini 3.7 FlashGoogle · Closed

85.5%

Claude Opus 5Anthropic · Closed

84.7%

Gemini 3.5 FlashGoogle · Closed

84.3%

GPT-5.6 SolOpenAI · Closed

83.4%

GPT-6 SolOpenAI · Closed

83.3%

Gemini 3.6 FlashGoogle · Closed

83.2%

Qwen3.8 Max PreviewAlibaba · Closed

82.8%

Gemini 3.1 ProGoogle · Closed

82.4%

GPT-5.6 TerraOpenAI · Closed

80.7%

Kimi K3Moonshot AI · Closed

80.5%

Qwen3.7 PlusAlibaba · Closed

80.5%

Muse SparkMeta · Closed

80.5%

Grok 4.5xAI · Closed

80.4%

Gemini 3 ProGoogle · Closed

80.2%

GPT-5.5OpenAI · Closed

79.9%

Qwen3.8-Flash-NextAlibaba · Open weight

79.8%

Kimi K2.6Moonshot AI · Open weight

79.4%

Apodex 1.1Apodex · Closed

79.2%

Apodex 1.1 MiniApodex · Open weight

79.2%

Ling 3.0 Flash VLInclusionAI · Open weight

79.0%

Gemini 3.5 Flash-LiteGoogle · Closed

79.0%

Claude Opus 4.7 (Adaptive)Anthropic · Closed

78.8%

GPT-5.6 LunaOpenAI · Closed

78.6%

MiniMax M3MiniMax · Open weight

78.6%

Gemini 3 FlashGoogle · Closed

78.6%

GPT-5.3 CodexOpenAI · Closed

78.5%

GPT-5.3-Codex-SparkOpenAI · Closed

78.5%

GPT-5.4OpenAI · Closed

78.4%

Grok 4.3xAI · Closed

78.1%

Qwen3.6 PlusAlibaba · Closed

78.0%

Claude Sonnet 5Anthropic · Closed

77.3%

DeepSeek V4.1 FlashDeepSeek · Open weight

77.0%

Step 5 PreviewStepFun · Closed

76.4%

Claude Opus 4.7Anthropic · Closed

76.4%

GPT-5.2-CodexOpenAI · Closed

76.3%

Qwen3.8-27BAlibaba · Open weight

76.3%

GPT-5.1OpenAI · Closed

75.5%

GPT-6 LunaOpenAI · Closed

75.5%

Kimi K2.5Moonshot AI · Open weight

75.4%

Kimi K2.5 (Reasoning)Moonshot AI · Closed

75.4%

Claude Opus 4.6 (Adaptive)Anthropic · Closed

75.4%

Step 3.7 FlashStepFun · Open weight

75.3%

Qwen3.5-27BAlibaba · Open weight

75.0%

Qwen3.5-122B-A10BAlibaba · Open weight

75.0%

Qwen3.6-35B-A3BAlibaba · Open weight

75.0%

Gemini 2.5 ProGoogle · Closed

74.9%

Qwen3.6-27BAlibaba · Open weight

74.6%

Muse Glimmer 30BMeta · Open weight

74.3%

GPT-5 (medium)OpenAI · Closed

74.3%

GPT-5 (high)OpenAI · Closed

74.2%

Inkling-SmallThinking Machines Lab · Open weight

74.0%

Claude Opus 4.5 ThinkingAnthropic · Closed

74.0%

InklingThinking Machines Lab · Open weight

73.5%

Gemma 4 31BGoogle · Open weight

73.4%

GPT-5.4 miniOpenAI · Closed

73.3%

GLM-5V-TurboZ.AI · Closed

72.8%

Qwen3.5-35B-A3BAlibaba · Open weight

72.7%

Claude Opus 4.6Anthropic · Closed

72.5%

GPT-5.1-CodexOpenAI · Closed

72.5%

GPT-5.1-Codex-MaxOpenAI · Closed

72.5%

Claude Opus 4.5Anthropic · Closed

71.2%

Claude Sonnet 4.6Anthropic · Closed

70.6%

o3OpenAI · Closed

70.1%

GPT-5 miniOpenAI · Closed

70.1%

MiMo-V2-OmniXiaomi · Closed

69.9%

Gemma 4 12BGoogle · Open weight

69.7%

Gemma 4 26B A4BGoogle · Open weight

69.2%

Grok 4xAI · Closed

68.8%

Claude 4.1 Opus ThinkingAnthropic · Closed

67.9%

Gemini 2.5 FlashGoogle · Closed

65.5%

GPT-5.4 nanoOpenAI · Closed

65.4%

Mistral Medium 3.5 128BMistral · Open weight

64.9%

Grok 4.1 Fast (Reasoning)xAI · Closed

63.3%

Command A+Cohere · Open weight

63.2%

Claude 4 SonnetAnthropic · Closed

62.4%

Llama 4 MaverickMeta · Open weight

62.1%

Grok 4 Fast (Reasoning)xAI · Closed

61.8%

GPT-4.1OpenAI · Closed

61.2%

GPT-5 nanoOpenAI · Closed

61.0%

Qwen3-Omni-30B-A3B-ThinkingAlibaba · Open weight

60.2%

GPT-4.1 miniOpenAI · Closed

58.7%

Mistral Small 4Mistral · Open weight

56.8%

Mistral Small 4 (Reasoning)Mistral · Open weight

56.8%

Mistral Large 3Mistral · Closed

55.7%

Qwen3-Omni-30B-A3B-InstructAlibaba · Open weight

55.5%

Gemini 1.5 ProGoogle · Closed

55.0%

Nemotron 3 Nano Omni 30B A3BNVIDIA · Open weight

53.2%

Mistral Medium 3Mistral · Closed

53.0%

Llama 4 ScoutMeta · Open weight

52.9%

Qwen3.5 397BAlibaba · Open weight

52.7%

Qwen3.5 397B (Reasoning)Alibaba · Open weight

52.7%

Gemma 4 E4BGoogle · Open weight

51.4%

Ministral 3 14B (Reasoning)Mistral · Open weight

49.8%

Ministral 3 14BMistral · Open weight

49.8%

Grok 4.1 FastxAI · Closed

48.4%

Gemma 3 27BGoogle · Open weight

48.0%

100

Ministral 3 8B (Reasoning)Mistral · Open weight

46.0%

101

Ministral 3 8BMistral · Open weight

46.0%

102

Gemma 4 E2BGoogle · Open weight

44.6%

103

Nova ProAmazon · Closed

44.3%

104

GPT-4o miniOpenAI · Closed

41.5%

105

GPT-4.1 nanoOpenAI · Closed

40.1%

106

Ministral 3 3B (Reasoning)Mistral · Open weight

38.1%

107

Ministral 3 3BMistral · Open weight

38.1%

108

Claude 3 HaikuAnthropic · Closed

30.8%

109

LFM2.5-VL-1.6B-ExtractLiquidAI · Open weight

26.5%

110

Phi-4 Multimodal InstructMicrosoft · Open weight

14.5%

According to BenchLM.ai, Claude Opus 5.5 leads the AA-MMMU-Pro benchmark with a score of 87.7%, followed by GPT-6 Astra (86.9%) and Gemini 3.8 Flash (85.6%). The top models are clustered within 2.1 points, suggesting this benchmark is nearing saturation for frontier models.

110 models have been evaluated on AA-MMMU-Pro. The benchmark falls in the Multimodal & Grounded category. The Multimodal & Grounded leaderboard ranks models by a weighted category score, and AA-MMMU-Pro contributes 15% of it. It does not enter the overall BenchAlign v5.7 ranking.

About AA-MMMU-Pro

Year

2026

Tasks

Multimodal academic reasoning

Format

Image + text question answering

Difficulty

Frontier multimodal

BenchLM stores the Artificial Analysis MMMU-Pro result separately from the core MMMU-Pro lane because Artificial Analysis controls the evaluation configuration.

Artificial Analysis MMMU-Pro Benchmark Leaderboard

Freshness and provenance

Version

AA-MMMU-Pro 2026

Refresh cadence

Quarterly

Staleness state

Current

Question availability

Public benchmark set

Current

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.

Questions

What does AA-MMMU-Pro measure?

An Artificial Analysis MMMU-Pro score.

Which model scores highest on AA-MMMU-Pro?

Claude Opus 5.5 by Anthropic currently leads with a score of 87.7% on AA-MMMU-Pro.

How many models are evaluated on AA-MMMU-Pro?

110 AI models have been evaluated on AA-MMMU-Pro on BenchLM.

Compare top models on AA-MMMU-Pro

Claude Opus 5.5 vs GPT-6 Astra GPT-6 Astra vs Gemini 3.8 Flash Gemini 3.8 Flash vs Gemini 3.7 Flash Gemini 3.7 Flash vs Claude Opus 5

Last updated: September 27, 2026 · BenchLM version AA-MMMU-Pro 2026

Know when it’s worth switching models

The model to choose, the cheaper alternative, and the release we would wait on.

Read a sample issue

Join 2,000+ readers.

One email each week. Unsubscribe anytime.