Subject · tested 27 Oct 2026

Best AI models for humanities & social science

Scores on the 9 humanities & social science questions in the suite (9 from MMLU-Pro). That is a small sample: the bars show each score's 95% interval, and overlapping bars mean the models can't be told apart.

Questions9of 80 in the suite
Top score9/9Qwen3.7 Plus, Muse Wonder, Llama 3.3 70B Instruct, Qwen3 Coder Next, Gemini 3.5 Flash Lite, Nemotron 3 Ultra, DeepSeek V4.1 Flash 1017, Qwen3.8 Max (0902)
Tied with the leader69overlapping 95% intervals
Median score77.8%of 72 models

Humanities & social science: score by model

9 questions · the bar shows each score's 95% interval

Humanities & social science leaderboard
#ModelCorrectScore95% interval
1Qwen3.7 PlusQwen9 / 9100.0%70–100%
1Muse WonderMeta · statistically tied with #19 / 9100.0%70–100%
1Llama 3.3 70B InstructMeta · statistically tied with #19 / 9100.0%70–100%
1Qwen3 Coder NextQwen · statistically tied with #19 / 9100.0%70–100%
1Gemini 3.5 Flash LiteGoogle · statistically tied with #19 / 9100.0%70–100%
1Nemotron 3 UltraNVIDIA · statistically tied with #19 / 9100.0%70–100%
1DeepSeek V4.1 Flash 1017DeepSeek · statistically tied with #19 / 9100.0%70–100%
1Qwen3.8 Max (0902)Qwen · statistically tied with #19 / 9100.0%70–100%
9InklingThinking Machines · statistically tied with #18 / 988.9%56–98%
9Step 5 PreviewStepFun · statistically tied with #18 / 988.9%56–98%
9Kimi K2.7 CodeMoonshotAI · statistically tied with #18 / 988.9%56–98%
9Qwen3.5 397B A17BQwen · statistically tied with #18 / 988.9%56–98%
9Claude Fable 5.1Anthropic · statistically tied with #18 / 988.9%56–98%
9GPT-6 AstraOpenAI · statistically tied with #18 / 988.9%56–98%
9Grok 4.7SpaceXAI · statistically tied with #18 / 988.9%56–98%
9o3OpenAI · statistically tied with #18 / 988.9%56–98%
9GPT-5.6 TerraOpenAI · statistically tied with #18 / 988.9%56–98%
9GPT-5.3-CodexOpenAI · statistically tied with #18 / 988.9%56–98%
9GPT-6.1 TerraOpenAI · statistically tied with #18 / 988.9%56–98%
9GPT-6.1 Terra ProOpenAI · statistically tied with #18 / 988.9%56–98%
9GLM 5.3 FlashZ.ai · statistically tied with #18 / 988.9%56–98%
9DeepSeek V4 Flash Vision ExpDeepSeek · statistically tied with #18 / 988.9%56–98%
9Qwen3.8 27BQwen · statistically tied with #18 / 988.9%56–98%
9Gemini 3.1 Pro PreviewGoogle · statistically tied with #18 / 988.9%56–98%
9DeepSeek V4 Pro 0813DeepSeek · statistically tied with #18 / 988.9%56–98%
9Muse Spark OpenMeta · statistically tied with #18 / 988.9%56–98%
9Claude Opus 5.5Anthropic · statistically tied with #18 / 988.9%56–98%
9Gemini 3.8 FlashGoogle · statistically tied with #18 / 988.9%56–98%
9GLM 5.3Z.ai · statistically tied with #18 / 988.9%56–98%
9GPT-6.1 SolOpenAI · statistically tied with #18 / 988.9%56–98%
9Gemini 4 ArgonGoogle · statistically tied with #18 / 988.9%56–98%
9GPT-6 SolOpenAI · statistically tied with #18 / 988.9%56–98%
9GPT-6 LunaOpenAI · statistically tied with #18 / 988.9%56–98%
9MiMo-V2.6-ProXiaomi · statistically tied with #18 / 988.9%56–98%
9Kimi K3.1 ProMoonshotAI · statistically tied with #18 / 988.9%56–98%
9Claude Sonnet 5.5Anthropic · statistically tied with #18 / 988.9%56–98%
37Mistral Medium 3.5Mistral · statistically tied with #17 / 977.8%45–94%
37Qwen3.8 2.4T A95BQwen · statistically tied with #17 / 977.8%45–94%
37Nemotron 3.5 Super VLNVIDIA · statistically tied with #17 / 977.8%45–94%
37Ling 3.1 FlashinclusionAI · statistically tied with #17 / 977.8%45–94%
37gpt-oss-20bOpenAI · statistically tied with #17 / 977.8%45–94%
37Ministral 3 8B 2512Mistral · statistically tied with #17 / 977.8%45–94%
37MiMo-V2.6-FlashXiaomi · statistically tied with #17 / 977.8%45–94%
37Qwen3.6 35B A3BQwen · statistically tied with #17 / 977.8%45–94%
37LongCat 2.0Meituan · statistically tied with #17 / 977.8%45–94%
37Inkling SmallThinking Machines · statistically tied with #17 / 977.8%45–94%
37GPT-5.4OpenAI · statistically tied with #17 / 977.8%45–94%
37Claude Haiku 5.5Anthropic · statistically tied with #17 / 977.8%45–94%
37MiniMax M3.1 FlashMiniMax · statistically tied with #17 / 977.8%45–94%
37Llama 4 MaverickMeta · statistically tied with #17 / 977.8%45–94%
51Command ACohere · statistically tied with #16 / 966.7%35–88%
51Llama 4 ScoutMeta · statistically tied with #16 / 966.7%35–88%
51Hermes 4 405BNous · statistically tied with #16 / 966.7%35–88%
51Mercury 2.5Inception · statistically tied with #16 / 966.7%35–88%
51Solar Mini 4Upstage · statistically tied with #16 / 966.7%35–88%
51Nemotron 3.5 LightningNVIDIA · statistically tied with #16 / 966.7%35–88%
51Hy3Tencent · statistically tied with #16 / 966.7%35–88%
51Solar Pro 4Upstage · statistically tied with #16 / 966.7%35–88%
51Mistral Small 4Mistral · statistically tied with #16 / 966.7%35–88%
51Trinity Large ThinkingArcee AI · statistically tied with #16 / 966.7%35–88%
51Mistral Large 4Mistral · statistically tied with #16 / 966.7%35–88%
51gpt-oss-120bOpenAI · statistically tied with #16 / 966.7%35–88%
63Gemma 4 31BGoogle · statistically tied with #15 / 955.6%27–81%
63Granite 4.0 MicroIBM · statistically tied with #15 / 955.6%27–81%
63Ministral 3 3B 2512Mistral · statistically tied with #15 / 955.6%27–81%
63Ministral 3 14B 2512Mistral · statistically tied with #15 / 955.6%27–81%
63MiniMax M3MiniMax · statistically tied with #15 / 955.6%27–81%
68Nova Micro 1.0Amazon · statistically tied with #14 / 944.4%19–73%
68Phi 4Microsoft · statistically tied with #14 / 944.4%19–73%
68Granite 4.2 8BIBM · statistically tied with #14 / 944.4%19–73%
71Gemma 4 26B A4BGoogle2 / 922.2%6–55%
72Reka Flash 3rekaai0 / 90.0%0–30%

How hard are these questions?

Questions by the share of 72 scored models that got them right

  1. 1No model right0 questions
  2. 2Under 25% of models1 question
  3. 325–50% of models0 questions
  4. 450–75% of models2 questions
  5. 575% or more, not all6 questions
  6. 6Every model right0 questions

Science →Engineering →Medicine →Economics →Law →

Humanities & social science: frequently asked questions

Which AI model is best at humanities & social science?
Qwen3.7 Plus and Muse Wonder and Llama 3.3 70B Instruct and Qwen3 Coder Next and Gemini 3.5 Flash Lite and Nemotron 3 Ultra and DeepSeek V4.1 Flash 1017 and Qwen3.8 Max (0902) answered the most humanities & social science questions correctly: 9 of 9. With only 9 questions, 69 other models are statistically tied with the leader, so treat this ranking as a rough guide.
Where do the humanities & social science questions come from?
9 from MMLU-Pro (MIT). Every question is multiple choice with ten options and one correct answer.
How hard are the humanities & social science questions?
0 of the 9 questions were answered correctly by none of the 72 scored models, and 0 by all of them.