Share promptsSign inCreate team library
← All benchmarks

Reasoning & knowledge AI benchmarks

Expert exams, graduate science, contest maths and broad knowledge. 4 public leaderboards below.

Humanity's Last Exam

2,500 expert-written, closed-ended questions across maths, science and the humanities, designed to be hard for frontier models.

Run by Scale AI and Center for AI Safety. Leaderboard date not stated · retrieved 21 Oct 2026. Showing top 20 of 55 models listed.

  • Anthropic
  • OpenAI
  • Google
  • Meta
  • Moonshot
Humanity's Last Exam: Calibration error (lower is better) vs AccuracyScatter plot of 18 models. Leader on Calibration error (lower is better): Fable 5.1 (xhigh) at 20%. Full values in the table below.10%20%30%40%50%60%0%20%40%60%80%Accuracy →↑ Calibration error (lower is better) (axis flipped so better is up)o3 (high) (April 2025) (OpenAI): Accuracy 20.32%, Calibration error (lower is better) 34%Ogemini-2.5-pro-preview-06-05 (Google): Accuracy 21.64%, Calibration error (lower is better) 72%Ggpt-5.1-thinking (OpenAI): Accuracy 23.68%, Calibration error (lower is better) 55%Okimi-k2.5 (Moonshot): Accuracy 24.37%, Calibration error (lower is better) 67%Kclaude-opus-4-5-20251101-thinking (Anthropic): Accuracy 25.20%, Calibration error (lower is better) 55%Agpt-5-2025-08-07 (OpenAI): Accuracy 25.32%, Calibration error (lower is better) 50%Ogpt-5.2-2025-12-11 (OpenAI): Accuracy 27.80%, Calibration error (lower is better) 45%Ogpt-5-pro-2025-10-06 (OpenAI): Accuracy 31.64%, Calibration error (lower is better) 49%Oclaude-opus-4-6-thinking-max (Anthropic): Accuracy 34.44%, Calibration error (lower is better) 46%Aclaude-opus-4-7 (Anthropic): Accuracy 36.20%, Calibration error (lower is better) 47%Agpt-5.4-2026-03-05 (xhigh thinking) (OpenAI): Accuracy 36.24%, Calibration error (lower is better) 42%Ogemini-3-pro-preview (Google): Accuracy 37.52%, Calibration error (lower is better) 57%GMuse Spark (Meta): Accuracy 40.56%, Calibration error (lower is better) 50%Mgpt-5.4-pro-2026-03-05 (OpenAI): Accuracy 44.32%, Calibration error (lower is better) 38%OGemini 3.8 Flash (Google): Accuracy 44.52%, Calibration error (lower is better) 51%Ggemini-3.1-pro-preview (thinking high) (Google): Accuracy 46.44%, Calibration error (lower is better) 51%GFable 5.1 (xhigh) (Anthropic): Accuracy 46.50%, Calibration error (lower is better) 20%AGPT 6 Astra (OpenAI): Accuracy 54.80%, Calibration error (lower is better) 39%OGPT 6 AstraFable 5.1o3 (high)gpt-5.4-pro-2026-03-05
Humanity's Last Exam: top 15 by AccuracyRanked bars. Leader: GPT 6 Astra at 54.80%. Full values in the table below.0%20%40%60%80%100%GPT 6 Astra (OpenAI): 54.80%OGPT 6 Astra54.80%Fable 5.1 (xhigh) (Anthropic): 46.50%AFable 5.1 (xhigh)46.50%gemini-3.1-pro-preview (thinking high) (Google): 46.44%Ggemini-3.1-pro-preview (think…46.44%Sonnet 5.5 (xhigh) (Anthropic): 46.12%ASonnet 5.5 (xhigh)46.12%Gemini 3.8 Flash (Google): 44.52%GGemini 3.8 Flash44.52%gpt-5.4-pro-2026-03-05 (OpenAI): 44.32%Ogpt-5.4-pro-2026-03-0544.32%Muse Spark (Meta): 40.56%MMuse Spark40.56%gemini-3-pro-preview (Google): 37.52%Ggemini-3-pro-preview37.52%gpt-5.4-2026-03-05 (xhigh thinking) (OpenAI): 36.24%Ogpt-5.4-2026-03-05 (xhigh thi…36.24%claude-opus-4-7 (Anthropic): 36.20%Aclaude-opus-4-736.20%claude-opus-4-6-thinking-max (Anthropic): 34.44%Aclaude-opus-4-6-thinking-max34.44%gpt-5-pro-2025-10-06 (OpenAI): 31.64%Ogpt-5-pro-2025-10-0631.64%Claude Haiku 5.5 (Anthropic): 30.12%AClaude Haiku 5.530.12%gpt-5.2-2025-12-11 (OpenAI): 27.80%Ogpt-5.2-2025-12-1127.80%gpt-5-2025-08-07 (OpenAI): 25.32%Ogpt-5-2025-08-0725.32%
Accuracy
Share of public HLE questions answered correctly (±95% confidence interval in the table).
Calibration error (lower is better)
RMS gap between the model’s stated confidence and its actual accuracy. High means overconfident.

Model names as listed by Scale. Two models have no published calibration error, so they appear in the table and ranked view only.

Humanity's Last Exam scores
RankModelProviderAccuracyCalibration error (lower is better)±
1 GPT 6 AstraOpenAI54.80%39%±1.94
2 Fable 5.1 (xhigh)Anthropic46.50%20%±2
3 gemini-3.1-pro-preview (thinking high)Google46.44%51%±1.96
4 Sonnet 5.5 (xhigh)Anthropic46.12%–±1.95
5 Gemini 3.8 FlashGoogle44.52%51%±1.96
6 gpt-5.4-pro-2026-03-05OpenAI44.32%38%±1.95
7 Muse SparkMeta40.56%50%±1.92
8 gemini-3-pro-previewGoogle37.52%57%±1.9
9 gpt-5.4-2026-03-05 (xhigh thinking)OpenAI36.24%42%±1.88
10 claude-opus-4-7Anthropic36.20%47%±1.88
Show 10 more rows
RankModelProviderAccuracyCalibration error (lower is better)±
11 claude-opus-4-6-thinking-maxAnthropic34.44%46%±1.86
12 gpt-5-pro-2025-10-06OpenAI31.64%49%±1.82
13 Claude Haiku 5.5Anthropic30.12%–±1.83
14 gpt-5.2-2025-12-11OpenAI27.80%45%±1.76
15 gpt-5-2025-08-07OpenAI25.32%50%±1.7
16 claude-opus-4-5-20251101-thinkingAnthropic25.20%55%±1.7
17 kimi-k2.5Moonshot24.37%67%±1.81
18 gpt-5.1-thinkingOpenAI23.68%55%±1.67
19 gemini-2.5-pro-preview-06-05Google21.64%72%±1.61
20 o3 (high) (April 2025)OpenAI20.32%34%±1.58

Source: Scale AI and Center for AI Safety leaderboard ↗

Top model here: GPT 6 Astra (OpenAI). Try prompts for ChatGPT →

GPQA Diamond

Graduate-level biology, physics and chemistry questions written to be “Google-proof”.

Run by Vals AI. Leaderboard as of 1 Sep 2026 · retrieved 21 Oct 2026. Showing top 25 of 138 models listed.

  • Anthropic
  • OpenAI
  • Google
  • xAI
  • Meta
  • DeepSeek
  • Alibaba
  • Moonshot
  • MiniMax
GPQA Diamond: top 15 by AccuracyRanked bars. Leader: Gemini 3.1 Pro Preview (02/26) at 95.45%. Full values in the table below.0%20%40%60%80%100%Gemini 3.1 Pro Preview (02/26) (Google): 95.45%GGemini 3.1 Pro Preview (02/26)95.45%GPT-5.6 Sol (OpenAI): 95.20%OGPT-5.6 Sol95.20%Grok 4.6 (xAI): 94.70%XGrok 4.694.70%Gemini 3.8 Flash (Google): 94.44%GGemini 3.8 Flash94.44%Gemini 3.7 Flash (Google): 93.94%GGemini 3.7 Flash93.94%Qwen 3.8 Max (Alibaba): 93.69%QQwen 3.8 Max93.69%Gemini 3.6 Flash (Google): 93.43%GGemini 3.6 Flash93.43%Claude Opus 5 (Anthropic): 93.43%AClaude Opus 593.43%Claude Fable 5.1 (Anthropic): 93.43%AClaude Fable 5.193.43%Claude Fable 5 (Anthropic): 93.18%AClaude Fable 593.18%GPT 5.5 (OpenAI): 93.18%OGPT 5.593.18%Kimi K3 (Moonshot): 92.93%KKimi K392.93%Grok 4.5 (xAI): 92.93%XGrok 4.592.93%Gemini 3.5 Flash (Google): 92.68%GGemini 3.5 Flash92.68%MiniMax-M3 (MiniMax): 92.68%MMMiniMax-M392.68%
Accuracy
Share of multiple-choice questions answered correctly.
GPQA Diamond scores
RankModelProviderAccuracyPrice in / out ($/1M tokens)
1 Gemini 3.1 Pro Preview (02/26)Google95.45%$2 / $12
2 GPT-5.6 SolOpenAI95.20%$4 / $20
3 Grok 4.6xAI94.70%$2 / $6
4 Gemini 3.8 FlashGoogle94.44%$1.5 / $7.5
5 Gemini 3.7 FlashGoogle93.94%$1.5 / $7.5
6 Qwen 3.8 MaxAlibaba93.69%$2 / $6
7 Gemini 3.6 FlashGoogle93.43%$1.5 / $7.5
8 Claude Opus 5Anthropic93.43%$5 / $25
9 Claude Fable 5.1Anthropic93.43%$10 / $50
10 Claude Fable 5Anthropic93.18%$10 / $50
Show 15 more rows
RankModelProviderAccuracyPrice in / out ($/1M tokens)
11 GPT 5.5OpenAI93.18%$5 / $30
12 Kimi K3Moonshot92.93%$3 / $15
13 Grok 4.5xAI92.93%$2 / $6
14 Gemini 3.5 FlashGoogle92.68%$1.5 / $9
15 MiniMax-M3MiniMax92.68%$0.6 / $2.4
16 Claude Opus 4.8Anthropic92.42%$5 / $25
17 DeepSeek V4 Pro 0813DeepSeek92.42%$1.32 / $3.96
18 GPT-5.6 LunaOpenAI91.67%$0.2 / $1.2
19 Gemini 3 Pro (11/25)Google91.67%$2 / $12
20 GPT 5.2OpenAI91.67%$1.75 / $14
21 GPT 5.4 (xhigh)OpenAI91.67%$2.5 / $15
22 Grok 4.3xAI91.41%$1.25 / $2.5
23 Muse Spark 1.1Meta91.16%$1.25 / $4.25
24 GPT-5.6 TerraOpenAI90.91%$2 / $12
25 Qwen 3.7 MaxAlibaba90.15%$2.5 / $7.5

Source: Vals AI leaderboard ↗

Top model here: Gemini 3.1 Pro Preview (02/26) (Google). Try prompts for Gemini →

AIME

Problems from the American Invitational Mathematics Examination, a hard high-school maths contest.

Run by Vals AI. Leaderboard as of 16 Apr 2026 · retrieved 21 Oct 2026. Showing top 25 of 96 models listed.

  • Anthropic
  • OpenAI
  • Google
  • xAI
  • Meta
  • Alibaba
  • Z.ai
  • Moonshot
AIME: top 15 by AccuracyRanked bars. Leader: Gemini 3.1 Pro Preview (02/26) at 98.13%. Full values in the table below.0%20%40%60%80%100%Gemini 3.1 Pro Preview (02/26) (Google): 98.13%GGemini 3.1 Pro Preview (02/26)98.13%GPT 5.2 (OpenAI): 96.88%OGPT 5.296.88%Muse Spark (Meta): 96.88%MMuse Spark96.88%Gemini 3 Pro (11/25) (Google): 96.68%GGemini 3 Pro (11/25)96.68%GPT 5.4 (xhigh) (OpenAI): 96.67%OGPT 5.4 (xhigh)96.67%Grok 4.20 (Reasoning) (xAI): 96.46%XGrok 4.20 (Reasoning)96.46%Claude Opus 4.7 (Anthropic): 96.25%AClaude Opus 4.796.25%Gemini 3 Flash (12/25) (Google): 95.63%GGemini 3 Flash (12/25)95.63%Claude Opus 4.6 (Thinking) (Anthropic): 95.63%AClaude Opus 4.6 (Thinking)95.63%GPT 5.4 Mini (OpenAI): 95.63%OGPT 5.4 Mini95.63%Kimi K2.5 (Moonshot): 95.63%KKimi K2.595.63%Claude Opus 4.5 (Thinking) (Anthropic): 95.42%AClaude Opus 4.5 (Thinking)95.42%Qwen 3.6 Plus (Alibaba): 94.58%QQwen 3.6 Plus94.58%GPT 5 (OpenAI): 93.37%OGPT 593.37%GPT 5.1 (OpenAI): 93.33%OGPT 5.193.33%
Accuracy
Share of problems answered correctly.
AIME scores
RankModelProviderAccuracyPrice in / out ($/1M tokens)
1 Gemini 3.1 Pro Preview (02/26)Google98.13%$2 / $12
2 GPT 5.2OpenAI96.88%$1.75 / $14
3 Muse SparkMeta96.88%N/A
4 Gemini 3 Pro (11/25)Google96.68%$2 / $12
5 GPT 5.4 (xhigh)OpenAI96.67%$2.5 / $15
6 Grok 4.20 (Reasoning)xAI96.46%$2 / $6
7 Claude Opus 4.7Anthropic96.25%$5 / $25
8 Gemini 3 Flash (12/25)Google95.63%$0.5 / $3
9 Claude Opus 4.6 (Thinking)Anthropic95.63%$5 / $25
10 GPT 5.4 MiniOpenAI95.63%$0.75 / $4.5
Show 15 more rows
RankModelProviderAccuracyPrice in / out ($/1M tokens)
11 Kimi K2.5Moonshot95.63%$0.6 / $3
12 Claude Opus 4.5 (Thinking)Anthropic95.42%$5 / $25
13 Qwen 3.6 PlusAlibaba94.58%$0.5 / $3
14 GPT 5OpenAI93.37%$1.25 / $10
15 GPT 5.1OpenAI93.33%$1.25 / $10
16 GLM 4.7Z.ai93.33%$0.6 / $2.2
17 GLM 4.6Z.ai92.71%$0.6 / $2.2
18 GPT OSS 120BOpenAI92.60%$0.15 / $0.6
19 Qwen 3.5 FlashAlibaba92.50%$0.1 / $0.4
20 Claude Sonnet 4.6Anthropic92.29%$3 / $15
21 Grok 4.1 Fast (Reasoning)xAI91.88%$0.2 / $0.5
22 GLM 5.1Z.ai91.88%$1 / $3.2
23 GLM 5Z.ai91.67%$1 / $3.2
24 GPT 5 MiniOpenAI91.46%$0.25 / $2
25 Grok 4 Fast (Reasoning)xAI91.25%$0.2 / $0.5

Source: Vals AI leaderboard ↗

Top model here: Gemini 3.1 Pro Preview (02/26) (Google). Try prompts for Gemini →

MMLU Pro

Harder, 10-option version of the classic MMLU knowledge test across 14 subjects.

Run by Vals AI. Leaderboard as of 1 Sep 2026 · retrieved 21 Oct 2026. Showing top 25 of 138 models listed.

  • Anthropic
  • OpenAI
  • Google
  • xAI
  • Meta
  • Alibaba
  • Moonshot
MMLU Pro: top 15 by AccuracyRanked bars. Leader: Claude Fable 5.1 at 92.38%. Full values in the table below.0%20%40%60%80%100%Claude Fable 5.1 (Anthropic): 92.38%AClaude Fable 5.192.38%Claude Opus 5 (Anthropic): 91.59%AClaude Opus 591.59%Claude Fable 5 (Anthropic): 91.50%AClaude Fable 591.50%Gemini 3.1 Pro Preview (02/26) (Google): 90.99%GGemini 3.1 Pro Preview (02/26)90.99%Gemini 3.8 Flash (Google): 90.22%GGemini 3.8 Flash90.22%Gemini 3.7 Flash (Google): 90.12%GGemini 3.7 Flash90.12%Gemini 3 Pro (11/25) (Google): 90.10%GGemini 3 Pro (11/25)90.10%Claude Opus 4.7 (Anthropic): 89.87%AClaude Opus 4.789.87%Claude Opus 4.8 (Anthropic): 89.58%AClaude Opus 4.889.58%Gemini 3.5 Flash (Google): 89.52%GGemini 3.5 Flash89.52%Grok 4.6 (xAI): 89.40%XGrok 4.689.40%Qwen 3.7 Max (Alibaba): 89.31%QQwen 3.7 Max89.31%Gemini 3.6 Flash (Google): 89.28%GGemini 3.6 Flash89.28%Grok 4.5 (xAI): 89.22%XGrok 4.589.22%Claude Opus 4.6 (Thinking) (Anthropic): 89.11%AClaude Opus 4.6 (Thinking)89.11%
Accuracy
Share of questions answered correctly.
MMLU Pro scores
RankModelProviderAccuracyPrice in / out ($/1M tokens)
1 Claude Fable 5.1Anthropic92.38%$10 / $50
2 Claude Opus 5Anthropic91.59%$5 / $25
3 Claude Fable 5Anthropic91.50%$10 / $50
4 Gemini 3.1 Pro Preview (02/26)Google90.99%$2 / $12
5 Gemini 3.8 FlashGoogle90.22%$1.5 / $7.5
6 Gemini 3.7 FlashGoogle90.12%$1.5 / $7.5
7 Gemini 3 Pro (11/25)Google90.10%$2 / $12
8 Claude Opus 4.7Anthropic89.87%$5 / $25
9 Claude Opus 4.8Anthropic89.58%$5 / $25
10 Gemini 3.5 FlashGoogle89.52%$1.5 / $9
Show 15 more rows
RankModelProviderAccuracyPrice in / out ($/1M tokens)
11 Grok 4.6xAI89.40%$2 / $6
12 Qwen 3.7 MaxAlibaba89.31%$2.5 / $7.5
13 Gemini 3.6 FlashGoogle89.28%$1.5 / $7.5
14 Grok 4.5xAI89.22%$2 / $6
15 Claude Opus 4.6 (Thinking)Anthropic89.11%$5 / $25
16 GPT-5.6 SolOpenAI89.10%$4 / $20
17 Muse Spark 1.1Meta88.73%$1.25 / $4.25
18 Qwen 3.8 MaxAlibaba88.60%$2 / $6
19 Gemini 3 Flash (12/25)Google88.59%$0.5 / $3
20 Muse Spark 1.2Meta88.28%$1.25 / $4.25
21 GPT 5.5OpenAI88.14%$5 / $30
22 Kimi K3Moonshot87.97%$3 / $15
23 Claude Opus 4.1 (Thinking)Anthropic87.92%$15 / $75
24 Qwen 3.6 PlusAlibaba87.67%$0.5 / $3
25 Kimi K2.6Moonshot87.57%$0.95 / $4

Source: Vals AI leaderboard ↗

Top model here: Claude Fable 5.1 (Anthropic). Try prompts for Claude →

ChatJudge does not run these tests. Scores are copied as published by each source, with a link and the date we read them, and only change when we refresh from the source. Different leaderboards use different settings, so compare models within one chart, not across charts. Spotted a mistake? Tell us.