Benchmark Almanac

SuperGPQA

Qwen 3.6 Max (Preview) 73.9% leads since April 2026
Gemini 3 Pro 73.8% leads since November 2025
Qwen 3.6 Max (Preview) 73.9% leads since April 2026
In use

The official leaderboard hasn’t scored a model released after GPT-5.2 Pro (December 2025). Labs report their own numbers for 27 newer models, such as Qwen 3.6 Max (Preview) at 73.9%.

26,529 graduate-level multiple-choice questions across 285 disciplines. Accuracy.

Made by
ByteDance Seed + M-A-P
Released
February 2025
Kind
Expert-written questions
Level
Expert
Scored
Checked automatically
Answers
Multiple choice
Tasks
All public
Measures
Capability
Language
English
In and out
Text → Text
Interaction
Single turn
Refresh
One fixed set
Run by
Anyone
Strongest model tested
GPT-5.2 Pro #32 on the capability index
Last in a launch post
October 2026
Last model added
October 2026
Scores refreshed
November 18th, 2026

In the authors’ words

Large language models (LLMs) have demonstrated remarkable proficiency in mainstream academic disciplines such as mathematics, physics, and computer science. However, human knowledge encompasses over 200 specialized disciplines, far exceeding the scope of existing benchmarks. The capabilities of LLMs in many of these specialized fields-particularly in light industry, agriculture, and service-oriented disciplines-remain inadequately evaluated. To address this gap, we present SuperGPQA, a comprehensive benchmark that evaluates graduate-level knowledge and reasoning capabilities across 285 disciplines. Our benchmark employs a novel Human-LLM collaborative filtering mechanism to eliminate trivial or ambiguous questions through iterative refinement based on both LLM responses and expert feedback. Our experimental results reveal significant room for improvement in the performance of current state-of-the-art LLMs across diverse knowledge domains (e.g., the reasoning-focused model DeepSeek-R1 achieved the highest accuracy of 61.82% on SuperGPQA), highlighting the considerable gap between current model capabilities and artificial general intelligence. Additionally, we present comprehensive insights from our management of a large-scale annotation process, involving over 80 expert annotators and an interactive Human-LLM collaborative system, offering valuable methodological guidance for future research initiatives of comparable scope.

Abstract of SuperGPQA: Scaling LLM Evaluation across 285 Graduate Disciplines (arXiv:2502.14739, CC0)
36
Models scored in lab cards
31
Launch posts list it, from 7 labs
14
System cards report it, from 8 labs

Qwen 3.6 Max (Preview) leads with 73.9%

Best score per model as labs report them in their own cards, drawn striped. Labs pick their own settings, so these compare less cleanly than independent runs.

Gemini 3 Pro leads with 73.8% across all independent runs

Best score per model across the official leaderboard, Epoch AI’s own runs and any board Epoch AI copied. Dotted bars are reruns by Epoch AI or another board, and each row names its source.

Qwen 3.6 Max (Preview) leads with 73.9%, claims included

Best score per model from every source, including labs’ own numbers and claims Epoch AI collected from reports and papers. Labs pick their own settings, so a claim can beat an independent run of the same model. Striped bars are claims, and dotted bars are reruns by Epoch AI or another board.

  1. 1 Qwen 3.6 Max (Preview) 73.9%
  2. 2 Qwen3.7-Max 73.6%
  3. 3 Qwen 3.6 Plus 71.6%
  4. 4 Qwen3.7-Plus 71.4%
  5. 5 Seed2.1 Pro 70.8%
  6. 6 Qwen3.5 397B-A17B 70.4%
  7. 7 Seed2.0 Lite (0428) 69.6%
  8. 8 Seed2.0 Pro 68.7%
  9. 9 Qwen3.5-122B-A10B 67.1%
  10. 10 Qwen3.5-Omni-Plus 66.4%
  11. 11 Qwen3.6 27B 66%
  12. 12 Qwen3.5-27B 65.6%
  13. 13 Qwen3-235B-A22B-Thinking Jul 2025 64.9%
  14. 14 Seed1.8 64.8%
  15. 15 Qwen 3.6 35B-A3B 64.7%
  16. 16 Qwen3-VL-235B-A22B 64.3%
  17. 17 Qwen3.5-35B-A3B 63.4%
  18. 18 Seed-Thinking-v1.5 62.1%
  19. 19 MiniCPM-V-4.7-35B-A3B 61.2%
  20. 20 Qwen3-Next-80B-A3B 60.8%
  21. 21 Ling-2.6-1T 58.3%
  22. 22 Qwen3.5-9B 58.2%
  23. 23 Qwen3-Coder-Next 57.5%
  24. 24 Kimi K2 Jul 2025 57.2%
  25. 25 seed-oss-36b-instruct 55.7%
  26. 26 DeepSeek-V4-Pro 53.9%
  27. 27 DeepSeek V4.1 Flash 53.1%
  28. 28 DeepSeek-V4.1-Flash Base 53.1%
  29. 29 Qwen3.5-4B 52.9%
  30. 30 GLM-Z1-32B-0414 52.6%
  31. 31 STEP3-VL-10B 50.4%
  32. 32 Nemotron 3.5 Super 120B-A12B Base 47.6%
  33. 33 MiMo-7B-RL 40.5%
  34. 34 Qwen3.5-2B 37.5%
  35. 35 Nemotron 3.5 Lightning 30B-A3B Base 33.2%
  36. 36 Qwen3.5-0.8B 21.3%

Source: Labs’ model cards

Source: Official SuperGPQA leaderboard

94 models. Scroll sideways for the rest.

  1. 73.9 Qwen 3.6 Max (Preview)
  2. 73.8 Gemini 3 Pro
  3. 73.6 Qwen3.7-Max
  4. 71.6 Qwen 3.6 Plus
  5. 71.4 Qwen3.7-Plus
  6. 70.8 Seed2.1 Pro
  7. 70.4 Qwen3.5 397B-A17B
  8. 69.6 Seed2.0 Lite (0428)
  9. 68.7 Seed2.0 Pro
  10. 67.1 GPT-5.2 Pro
  11. 67.1 Qwen3.5-122B-A10B
  12. 66.4 Qwen3.5-Omni-Plus
  13. 66.3 GPT-5.1
  14. 66.0 Qwen3.6 27B
  15. 66.0 Gemini 2.5 Pro
  16. 65.6 GPT-5
  17. 65.6 Qwen3.5-27B
  18. 64.9 Qwen3-235B-A22B-Thinking
  19. 64.8 Seed1.8
  20. 64.7 Qwen 3.6 35B-A3B
  21. 64.3 Qwen3-VL-235B-A22B
  22. 63.4 Qwen3.5-35B-A3B
  23. 63.0 Grok 4
  24. 62.1 Seed-Thinking-v1.5
  25. 61.8 DeepSeek-R1
  26. 61.4 Claude Opus 4
  27. 61.3 GPT-5.2
  28. 61.2 MiniCPM-V-4.7-35B-A3B
  29. 60.8 Qwen3-Next-80B-A3B
  30. 60.2 o1
  31. 60.2 DeepSeek-R1-Zero
  32. 59.3 DeepSeek-V3.1
  33. 59.2 Gemini 2.5 Flash
  34. 58.8 GPT-5 mini
  35. 58.3 Ling-2.6-1T
  36. 58.2 Qwen3.5-9B
  37. 58.1 Kimi K2
  38. 57.5 Qwen3-Coder-Next
  39. 57.4 GPT-5 Chat
  40. 57.0 Claude Sonnet 4
  41. 56.4 GPT-5.2 Instant
  42. 55.7 seed-oss-36b-instruct
  43. 55.2 o3-mini
  44. 53.9 DeepSeek-V4-Pro
  45. 53.1 DeepSeek V4.1 Flash
  46. 53.1 DeepSeek-V4.1-Flash Base
  47. 52.9 Qwen3.5-4B
  48. 52.6 GLM-Z1-32B-0414
  49. 51.9 gpt-oss-120b
  50. 50.9 Doubao-1.5-pro
  51. 50.4 STEP3-VL-10B
  52. 50.1 Qwen2.5-Max
  53. 49.4 Qwen3-235B-A22B
  54. 48.9 GPT-5 nano
  55. 48.2 Claude 3.5 Sonnet (October 2024)
  56. 47.7 Gemini 2.0 Flash
  57. 47.6 Nemotron 3.5 Super 120B-A12B Base
  58. 47.4 DeepSeek-V3
  59. 45.2 o1-mini
  60. 45.1 MiniMax-Text-01
  61. 44.9 gpt-oss-20b
  62. 44.4 GPT-4o
  63. 43.6 QwQ-32B-Preview
  64. 43.1 Llama 3.1-405B
  65. 41.6 GPT-4o
  66. 40.8 Qwen2.5-72B
  67. 40.6 Mistral Large 2
  68. 40.5 MiMo-7B-RL
  69. 39.8 GPT-4o
  70. 38.8 Qwen2.5-32B
  71. 37.7 Llama 3.3 70B Instruct
  72. 37.6 Phi-4
  73. 37.5 Qwen3.5-2B
  74. 35.1 Qwen2.5-14B
  75. 34.9 Llama 3.1-70B
  76. 33.2 Nemotron 3.5 Lightning 30B-A3B Base
  77. 29.2 Mixtral 8x22B
  78. 28.8 Qwen2.5-7B
  79. 27.4 Gemma 2 27B
  80. 26.0 Yi-1.5-34B
  81. 25.9 Mistral Small v24.09
  82. 24.0 Gemma 2 9B
  83. 23.3 Qwen2.5-3B
  84. 23.2 Yi-1.5-9B-Chat
  85. 22.1 Mixtral 8x7B
  86. 21.3 Qwen3.5-0.8B
  87. 20.8 Granite-3.1-8B-Instruct
  88. 20.5 Llama 3.1-8B
  89. 19.2 Yi-1.5-6B-Chat
  90. 18.8 Qwen2.5-1.5B
  91. 18.6 Gemma 2 2B IT
  92. 17.9 Granite-3.1-2B-Instruct
  93. 17.8 Mistral 7B v0.3
  94. 10.8 Qwen2.5-0.5B
  1. 1 Qwen 3.6 Max (Preview) 73.9%
  2. 2 Gemini 3 Pro 73.8%
  3. 3 Qwen3.7-Max 73.6%
  4. 4 Qwen 3.6 Plus 71.6%
  5. 5 Qwen3.7-Plus 71.4%
  6. 6 Seed2.1 Pro 70.8%
  7. 7 Qwen3.5 397B-A17B 70.4%
  8. 8 Seed2.0 Lite (0428) 69.6%
  9. 9 Seed2.0 Pro 68.7%
  10. 10 GPT-5.2 Pro 67.1%
  11. 11 Qwen3.5-122B-A10B 67.1%
  12. 12 Qwen3.5-Omni-Plus 66.4%
  13. 13 GPT-5.1 thinking 66.3%
  14. 14 Qwen3.6 27B 66%
  15. 15 Gemini 2.5 Pro Jun 2025, five-shot 66%
  16. 16 GPT-5 65.6%
  17. 17 Qwen3.5-27B 65.6%
  18. 18 Qwen3-235B-A22B-Thinking Jul 2025 64.9%
  19. 19 Seed1.8 64.8%
  20. 20 Qwen 3.6 35B-A3B 64.7%
  21. 21 Qwen3-VL-235B-A22B 64.3%
  22. 22 Qwen3.5-35B-A3B 63.4%
  23. 23 Grok 4 63%
  24. 24 Seed-Thinking-v1.5 62.1%
  25. 25 DeepSeek-R1 61.8%
  26. 26 Claude Opus 4 61.4%
  27. 27 GPT-5.2 61.3%
  28. 28 MiniCPM-V-4.7-35B-A3B 61.2%
  29. 29 Qwen3-Next-80B-A3B 60.8%
  30. 30 o1 60.2%
  31. 31 DeepSeek-R1-Zero 60.2%
  32. 32 DeepSeek-V3.1 59.3%
  33. 33 Gemini 2.5 Flash Jun 2025, five-shot 59.2%
  34. 34 GPT-5 mini 58.8%
  35. 35 Ling-2.6-1T 58.3%
  36. 36 Qwen3.5-9B 58.2%
  37. 37 Kimi K2 Jul 2025 58.1%
  38. 38 Qwen3-Coder-Next 57.5%
  39. 39 GPT-5 Chat 57.4%
  40. 40 Claude Sonnet 4 57%
  41. 41 GPT-5.2 Instant 56.4%
  42. 42 seed-oss-36b-instruct 55.7%
  43. 43 o3-mini high 55.2%
  44. 44 DeepSeek-V4-Pro 53.9%
  45. 45 DeepSeek V4.1 Flash 53.1%
  46. 46 DeepSeek-V4.1-Flash Base 53.1%
  47. 47 Qwen3.5-4B 52.9%
  48. 48 GLM-Z1-32B-0414 52.6%
  49. 49 gpt-oss-120b 51.9%
  50. 50 Doubao-1.5-pro 50.9%
  51. 51 STEP3-VL-10B 50.4%
  52. 52 Qwen2.5-Max 50.1%
  53. 53 Qwen3-235B-A22B 49.4%
  54. 54 GPT-5 nano 48.9%
  55. 55 Claude 3.5 Sonnet (October 2024) 48.2%
  56. 56 Gemini 2.0 Flash Feb 2025 47.7%
  57. 57 Nemotron 3.5 Super 120B-A12B Base 47.6%
  58. 58 DeepSeek-V3 47.4%
  59. 59 o1-mini 45.2%
  60. 60 MiniMax-Text-01 45.1%
  61. 61 gpt-oss-20b 44.9%
  62. 62 GPT-4o Nov 2024 44.4%
  63. 63 QwQ-32B-Preview 43.6%
  64. 64 Llama 3.1-405B 43.1%
  65. 65 GPT-4o Aug 2024 41.6%
  66. 66 Qwen2.5-72B 40.8%
  67. 67 Mistral Large 2 Nov 2024 40.6%
  68. 68 MiMo-7B-RL 40.5%
  69. 69 GPT-4o May 2024 39.8%
  70. 70 Qwen2.5-32B 38.8%
  71. 71 Llama 3.3 70B Instruct 37.7%
  72. 72 Phi-4 37.6%
  73. 73 Qwen3.5-2B 37.5%
  74. 74 Qwen2.5-14B 35.1%
  75. 75 Llama 3.1-70B 34.9%
  76. 76 Nemotron 3.5 Lightning 30B-A3B Base 33.2%
  77. 77 Mixtral 8x22B 29.2%
  78. 78 Qwen2.5-7B 28.8%
  79. 79 Gemma 2 27B 27.4%
  80. 80 Yi-1.5-34B 26%
  81. 81 Mistral Small v24.09 25.9%
  82. 82 Gemma 2 9B 24%
  83. 83 Qwen2.5-3B 23.3%
  84. 84 Yi-1.5-9B-Chat 23.2%
  85. 85 Mixtral 8x7B 22.1%
  86. 86 Qwen3.5-0.8B 21.3%
  87. 87 Granite-3.1-8B-Instruct 20.8%
  88. 88 Llama 3.1-8B 20.5%
  89. 89 Yi-1.5-6B-Chat 19.2%
  90. 90 Qwen2.5-1.5B 18.8%
  91. 91 Gemma 2 2B IT 18.6%
  92. 92 Granite-3.1-2B-Instruct 17.9%
  93. 93 Mistral 7B v0.3 17.8%
  94. 94 Qwen2.5-0.5B 10.8%

Sources: Labs’ model cards and official SuperGPQA leaderboard

#ModelScoreSettingDateSource
1 Qwen 3.6 Max (Preview) 73.9% Apr 2026 Lab model card
2 Qwen3.7-Max 73.6% May 2026 Lab model card
3 Qwen 3.6 Plus 71.6% Apr 2026 Lab model card
4 Qwen3.7-Plus 71.4% Jun 2026 Lab model card
5 Seed2.1 Pro 70.8% Jun 2026 Lab model card
6 Qwen3.5 397B-A17B 70.4% Feb 2026 Lab model card
7 Seed2.0 Lite (0428) 69.6% Apr 2026 Lab model card
8 Seed2.0 Pro 68.7% Feb 2026 Lab model card
9 Qwen3.5-122B-A10B 67.1% Feb 2026 Lab model card
10 Qwen3.5-Omni-Plus 66.4% Mar 2026 Lab model card
11 Qwen3.6 27B 66% Apr 2026 Lab model card
12 Qwen3.5-27B 65.6% Feb 2026 Lab model card
13 Qwen3-235B-A22B-Thinking Jul 2025 64.9% Jul 2025 Lab model card
14 Seed1.8 64.8% Dec 2025 Lab model card
15 Qwen 3.6 35B-A3B 64.7% Apr 2026 Lab model card
16 Qwen3-VL-235B-A22B 64.3% Sep 2025 Lab model card
17 Qwen3.5-35B-A3B 63.4% Feb 2026 Lab model card
18 Seed-Thinking-v1.5 62.1% Apr 2025 Lab model card
19 MiniCPM-V-4.7-35B-A3B 61.2% Oct 2026 Lab model card
20 Qwen3-Next-80B-A3B 60.8% Sep 2025 Lab model card
21 Ling-2.6-1T 58.3% Jun 2026 Lab model card
22 Qwen3.5-9B 58.2% Mar 2026 Lab model card
23 Qwen3-Coder-Next 57.5% Feb 2026 Lab model card
24 Kimi K2 Jul 2025 57.2% Jul 2025 Lab model card
25 seed-oss-36b-instruct 55.7% Aug 2025 Lab model card
26 DeepSeek-V4-Pro 53.9% Apr 2026 Lab model card
27 DeepSeek V4.1 Flash 53.1% Sep 2026 Lab model card
28 DeepSeek-V4.1-Flash Base 53.1% Sep 2026 Lab model card
29 Qwen3.5-4B 52.9% Mar 2026 Lab model card
30 GLM-Z1-32B-0414 52.6% Apr 2025 Lab model card
31 STEP3-VL-10B 50.4% Jan 2026 Lab model card
32 Nemotron 3.5 Super 120B-A12B Base 47.6% Oct 2026 Lab model card
33 MiMo-7B-RL 40.5% Apr 2025 Lab model card
34 Qwen3.5-2B 37.5% Mar 2026 Lab model card
35 Nemotron 3.5 Lightning 30B-A3B Base 33.2% Oct 2026 Lab model card
36 Qwen3.5-0.8B 21.3% Mar 2026 Lab model card
#ModelScoreSettingDateSource
1 Gemini 3 Pro 73.8% Nov 2025 Official leaderboard
2 GPT-5.2 Pro 67.1% Dec 2025 Official leaderboard
3 GPT-5.1 66.3% thinking Nov 2025 Official leaderboard
4 Gemini 2.5 Pro Jun 2025 66% five-shot Jun 2025 Official leaderboard
5 GPT-5 65.6% Aug 2025 Official leaderboard
6 Grok 4 63% Jul 2025 Official leaderboard
7 DeepSeek-R1 61.8% Jan 2025 Official leaderboard
8 Claude Opus 4 61.4% May 2025 Official leaderboard
9 GPT-5.2 61.3% Dec 2025 Official leaderboard
10 o1 60.2% Dec 2024 Official leaderboard
11 DeepSeek-R1-Zero 60.2% Jan 2025 Official leaderboard
12 DeepSeek-V3.1 59.3% Aug 2025 Official leaderboard
13 Gemini 2.5 Flash Jun 2025 59.2% five-shot Jun 2025 Official leaderboard
14 GPT-5 mini 58.8% Aug 2025 Official leaderboard
15 Kimi K2 Jul 2025 58.1% Jul 2025 Official leaderboard
16 GPT-5 Chat 57.4% Aug 2025 Official leaderboard
17 Claude Sonnet 4 57% May 2025 Official leaderboard
18 GPT-5.2 Instant 56.4% Dec 2025 Official leaderboard
19 o3-mini 55.2% high Jan 2025 Official leaderboard
20 gpt-oss-120b 51.9% Aug 2025 Official leaderboard
21 Doubao-1.5-pro 50.9% Jan 2025 Official leaderboard
22 Qwen2.5-Max 50.1% Jan 2025 Official leaderboard
23 Qwen3-235B-A22B 49.4% Apr 2025 Official leaderboard
24 GPT-5 nano 48.9% Aug 2025 Official leaderboard
25 Claude 3.5 Sonnet (October 2024) 48.2% Oct 2024 Official leaderboard
26 Gemini 2.0 Flash Feb 2025 47.7% Feb 2025 Official leaderboard
27 DeepSeek-V3 47.4% Dec 2024 Official leaderboard
28 o1-mini 45.2% Sep 2024 Official leaderboard
29 MiniMax-Text-01 45.1% Jan 2025 Official leaderboard
30 gpt-oss-20b 44.9% Aug 2025 Official leaderboard
31 GPT-4o Nov 2024 44.4% Nov 2024 Official leaderboard
32 QwQ-32B-Preview 43.6% Nov 2024 Official leaderboard
33 Llama 3.1-405B 43.1% Jul 2024 Official leaderboard
34 GPT-4o Aug 2024 41.6% Aug 2024 Official leaderboard
35 Qwen2.5-72B 40.8% Sep 2024 Official leaderboard
36 Mistral Large 2 Nov 2024 40.6% Nov 2024 Official leaderboard
37 GPT-4o May 2024 39.8% May 2024 Official leaderboard
38 Qwen2.5-32B 38.8% Sep 2024 Official leaderboard
39 Llama 3.3 70B Instruct 37.7% Dec 2024 Official leaderboard
40 Phi-4 37.6% Dec 2024 Official leaderboard
41 Qwen2.5-14B 35.1% Sep 2024 Official leaderboard
42 Llama 3.1-70B 34.9% Jul 2024 Official leaderboard
43 Mixtral 8x22B 29.2% Apr 2024 Official leaderboard
44 Qwen2.5-7B 28.8% Sep 2024 Official leaderboard
45 Gemma 2 27B 27.4% Jun 2024 Official leaderboard
46 Yi-1.5-34B 26% May 2024 Official leaderboard
47 Mistral Small v24.09 25.9% Sep 2024 Official leaderboard
48 Gemma 2 9B 24% Jun 2024 Official leaderboard
49 Qwen2.5-3B 23.3% Sep 2024 Official leaderboard
50 Yi-1.5-9B-Chat 23.2% May 2024 Official leaderboard
51 Mixtral 8x7B 22.1% Dec 2023 Official leaderboard
52 Granite-3.1-8B-Instruct 20.8% Dec 2024 Official leaderboard
53 Llama 3.1-8B 20.5% Jul 2024 Official leaderboard
54 Yi-1.5-6B-Chat 19.2% May 2024 Official leaderboard
55 Qwen2.5-1.5B 18.8% Sep 2024 Official leaderboard
56 Gemma 2 2B IT 18.6% Jul 2024 Official leaderboard
57 Granite-3.1-2B-Instruct 17.9% Dec 2024 Official leaderboard
58 Mistral 7B v0.3 17.8% Sep 2023 Official leaderboard
59 Qwen2.5-0.5B 10.8% Sep 2024 Official leaderboard
#ModelScoreSettingDateSource
1 Qwen 3.6 Max (Preview) 73.9% Apr 2026 Lab model card
2 Gemini 3 Pro 73.8% Nov 2025 Official leaderboard
3 Qwen3.7-Max 73.6% May 2026 Lab model card
4 Qwen 3.6 Plus 71.6% Apr 2026 Lab model card
5 Qwen3.7-Plus 71.4% Jun 2026 Lab model card
6 Seed2.1 Pro 70.8% Jun 2026 Lab model card
7 Qwen3.5 397B-A17B 70.4% Feb 2026 Lab model card
8 Seed2.0 Lite (0428) 69.6% Apr 2026 Lab model card
9 Seed2.0 Pro 68.7% Feb 2026 Lab model card
10 GPT-5.2 Pro 67.1% Dec 2025 Official leaderboard
11 Qwen3.5-122B-A10B 67.1% Feb 2026 Lab model card
12 Qwen3.5-Omni-Plus 66.4% Mar 2026 Lab model card
13 GPT-5.1 66.3% thinking Nov 2025 Official leaderboard
14 Qwen3.6 27B 66% Apr 2026 Lab model card
15 Gemini 2.5 Pro Jun 2025 66% five-shot Jun 2025 Official leaderboard
16 GPT-5 65.6% Aug 2025 Official leaderboard
17 Qwen3.5-27B 65.6% Feb 2026 Lab model card
18 Qwen3-235B-A22B-Thinking Jul 2025 64.9% Jul 2025 Lab model card
19 Seed1.8 64.8% Dec 2025 Lab model card
20 Qwen 3.6 35B-A3B 64.7% Apr 2026 Lab model card
21 Qwen3-VL-235B-A22B 64.3% Sep 2025 Lab model card
22 Qwen3.5-35B-A3B 63.4% Feb 2026 Lab model card
23 Grok 4 63% Jul 2025 Official leaderboard
24 Seed-Thinking-v1.5 62.1% Apr 2025 Lab model card
25 DeepSeek-R1 61.8% Jan 2025 Official leaderboard
26 Claude Opus 4 61.4% May 2025 Official leaderboard
27 GPT-5.2 61.3% Dec 2025 Official leaderboard
28 MiniCPM-V-4.7-35B-A3B 61.2% Oct 2026 Lab model card
29 Qwen3-Next-80B-A3B 60.8% Sep 2025 Lab model card
30 o1 60.2% Dec 2024 Official leaderboard
31 DeepSeek-R1-Zero 60.2% Jan 2025 Official leaderboard
32 DeepSeek-V3.1 59.3% Aug 2025 Official leaderboard
33 Gemini 2.5 Flash Jun 2025 59.2% five-shot Jun 2025 Official leaderboard
34 GPT-5 mini 58.8% Aug 2025 Official leaderboard
35 Ling-2.6-1T 58.3% Jun 2026 Lab model card
36 Qwen3.5-9B 58.2% Mar 2026 Lab model card
37 Kimi K2 Jul 2025 58.1% Jul 2025 Official leaderboard
38 Qwen3-Coder-Next 57.5% Feb 2026 Lab model card
39 GPT-5 Chat 57.4% Aug 2025 Official leaderboard
40 Claude Sonnet 4 57% May 2025 Official leaderboard
41 GPT-5.2 Instant 56.4% Dec 2025 Official leaderboard
42 seed-oss-36b-instruct 55.7% Aug 2025 Lab model card
43 o3-mini 55.2% high Jan 2025 Official leaderboard
44 DeepSeek-V4-Pro 53.9% Apr 2026 Lab model card
45 DeepSeek V4.1 Flash 53.1% Sep 2026 Lab model card
46 DeepSeek-V4.1-Flash Base 53.1% Sep 2026 Lab model card
47 Qwen3.5-4B 52.9% Mar 2026 Lab model card
48 GLM-Z1-32B-0414 52.6% Apr 2025 Lab model card
49 gpt-oss-120b 51.9% Aug 2025 Official leaderboard
50 Doubao-1.5-pro 50.9% Jan 2025 Official leaderboard
51 STEP3-VL-10B 50.4% Jan 2026 Lab model card
52 Qwen2.5-Max 50.1% Jan 2025 Official leaderboard
53 Qwen3-235B-A22B 49.4% Apr 2025 Official leaderboard
54 GPT-5 nano 48.9% Aug 2025 Official leaderboard
55 Claude 3.5 Sonnet (October 2024) 48.2% Oct 2024 Official leaderboard
56 Gemini 2.0 Flash Feb 2025 47.7% Feb 2025 Official leaderboard
57 Nemotron 3.5 Super 120B-A12B Base 47.6% Oct 2026 Lab model card
58 DeepSeek-V3 47.4% Dec 2024 Official leaderboard
59 o1-mini 45.2% Sep 2024 Official leaderboard
60 MiniMax-Text-01 45.1% Jan 2025 Official leaderboard
61 gpt-oss-20b 44.9% Aug 2025 Official leaderboard
62 GPT-4o Nov 2024 44.4% Nov 2024 Official leaderboard
63 QwQ-32B-Preview 43.6% Nov 2024 Official leaderboard
64 Llama 3.1-405B 43.1% Jul 2024 Official leaderboard
65 GPT-4o Aug 2024 41.6% Aug 2024 Official leaderboard
66 Qwen2.5-72B 40.8% Sep 2024 Official leaderboard
67 Mistral Large 2 Nov 2024 40.6% Nov 2024 Official leaderboard
68 MiMo-7B-RL 40.5% Apr 2025 Lab model card
69 GPT-4o May 2024 39.8% May 2024 Official leaderboard
70 Qwen2.5-32B 38.8% Sep 2024 Official leaderboard
71 Llama 3.3 70B Instruct 37.7% Dec 2024 Official leaderboard
72 Phi-4 37.6% Dec 2024 Official leaderboard
73 Qwen3.5-2B 37.5% Mar 2026 Lab model card
74 Qwen2.5-14B 35.1% Sep 2024 Official leaderboard
75 Llama 3.1-70B 34.9% Jul 2024 Official leaderboard
76 Nemotron 3.5 Lightning 30B-A3B Base 33.2% Oct 2026 Lab model card
77 Mixtral 8x22B 29.2% Apr 2024 Official leaderboard
78 Qwen2.5-7B 28.8% Sep 2024 Official leaderboard
79 Gemma 2 27B 27.4% Jun 2024 Official leaderboard
80 Yi-1.5-34B 26% May 2024 Official leaderboard
81 Mistral Small v24.09 25.9% Sep 2024 Official leaderboard
82 Gemma 2 9B 24% Jun 2024 Official leaderboard
83 Qwen2.5-3B 23.3% Sep 2024 Official leaderboard
84 Yi-1.5-9B-Chat 23.2% May 2024 Official leaderboard
85 Mixtral 8x7B 22.1% Dec 2023 Official leaderboard
86 Qwen3.5-0.8B 21.3% Mar 2026 Lab model card
87 Granite-3.1-8B-Instruct 20.8% Dec 2024 Official leaderboard
88 Llama 3.1-8B 20.5% Jul 2024 Official leaderboard
89 Yi-1.5-6B-Chat 19.2% May 2024 Official leaderboard
90 Qwen2.5-1.5B 18.8% Sep 2024 Official leaderboard
91 Gemma 2 2B IT 18.6% Jul 2024 Official leaderboard
92 Granite-3.1-2B-Instruct 17.9% Dec 2024 Official leaderboard
93 Mistral 7B v0.3 17.8% Sep 2023 Official leaderboard
94 Qwen2.5-0.5B 10.8% Sep 2024 Official leaderboard

The benchmark's own leaderboard has 90 entries, 59 of them models we track. Switch the scores above to rank them together.

Rankings by part

We read 3 parts from the benchmark’s own leaderboard: Hard, Middle and Easy. Each ranks models on one part of the test. Gemini 3 Pro leads every one of them. Arrows show how far a model moved from its overall rank among the models we track.

Hard

59 models we track have a score here.

  1. 1 Gemini 3 Pro 68%
  2. 2 GPT-5.2 Pro 63.6%
  3. 3 GPT-5.1 62.4%
  4. 4 Gemini 2.5 Pro Jun 2025 61.2%
  5. 5 GPT-5 59.9%
  6. 6 Grok 4 58.1%
  7. 7 DeepSeek-R1 56.9%
  8. 8 GPT-5 mini ↑6 56.9%
  9. 9 o3-mini ↑10 56.2%
  10. 10 Gemini 2.5 Flash Jun 2025 ↑3 56.1%

Scores at each effort level

Models the leaderboard ran at two or more effort levels. We don’t have enough cost data here to say what the extra effort costs.

Lowest effort to highest, light to dark
GPT-5.1
thinking
66.3%
Gemini 2.5 Pro
five-shot
66%
Claude Opus 4
five-shot
61.4%
Gemini 2.5 Flash
five-shot
59.2%
Kimi K2
five-shot
58.1%
Claude Sonnet 4
five-shot
57%
o3-mini
low medium high
55.2%
gpt-oss-120b
five-shot
51.9%
Qwen3-235B-A22B
five-shot
49.4%

Source: Official SuperGPQA leaderboard

ModelEffortHarnessScoreCost per taskOutput tokens
Gemini 3 Pro 73.8%
GPT-5.2 Pro 67.1%
GPT-5.1 thinking 66.3%
Gemini 2.5 Pro Jun 2025 five-shot 66%
GPT-5 65.6%
Gemini 2.5 Pro Jun 2025 64.5%
Grok 4 63%
DeepSeek-R1 61.8%
Claude Opus 4 61.4%
GPT-5.2 61.3%
o1 60.2%
DeepSeek-R1-Zero 60.2%
Claude Opus 4 five-shot 60.1%
DeepSeek-V3.1 59.3%
Gemini 2.5 Flash Jun 2025 five-shot 59.2%
Gemini 2.5 Flash Jun 2025 58.9%
GPT-5 mini 58.8%
Kimi K2 Jul 2025 58.1%
Kimi K2 Jul 2025 five-shot 57.7%
GPT-5 Chat 57.4%
Claude Sonnet 4 57%
GPT-5.2 Instant 56.4%
Claude Sonnet 4 five-shot 55.7%
o3-mini high 55.2%
o3-mini medium 52.7%
GPT-5.1 52.1%
gpt-oss-120b 51.9%
gpt-oss-120b five-shot 51.2%
Doubao-1.5-pro 50.9%
Qwen2.5-Max 50.1%
Qwen3-235B-A22B 49.4%
GPT-5 nano 48.9%
Claude 3.5 Sonnet (October 2024) 48.2%
o3-mini low 48%
Gemini 2.0 Flash Feb 2025 47.7%
o3-mini 47.5%
DeepSeek-V3 47.4%
Qwen3-235B-A22B five-shot 45.5%
o1-mini 45.2%
MiniMax-Text-01 45.1%
gpt-oss-20b 44.9%
GPT-4o Nov 2024 44.4%
QwQ-32B-Preview 43.6%
Llama 3.1-405B 43.1%
GPT-4o Aug 2024 41.6%
Qwen2.5-72B 40.8%
Mistral Large 2 Nov 2024 40.6%
GPT-4o May 2024 39.8%
Qwen2.5-32B 38.8%
Llama 3.3 70B Instruct 37.7%
Phi-4 37.6%
Qwen2.5-14B 35.1%
Llama 3.1-70B 34.9%
Mixtral 8x22B 29.2%
Qwen2.5-7B 28.8%
Gemma 2 27B 27.4%
Yi-1.5-34B 26%
Mistral Small v24.09 25.9%
Gemma 2 9B 24%
Qwen2.5-3B 23.3%
Yi-1.5-9B-Chat 23.2%
Mixtral 8x7B 22.1%
Granite-3.1-8B-Instruct 20.8%
Llama 3.1-8B 20.5%
Yi-1.5-6B-Chat 19.2%
Qwen2.5-1.5B 18.8%
Gemma 2 2B IT 18.6%
Granite-3.1-2B-Instruct 17.9%
Mistral 7B v0.3 17.8%
Qwen2.5-0.5B 10.8%

6 models have held number one in labs’ own numbers since its release in February 2025 4 models have held number one across all independent runs since its release in February 2025 6 models have held number one with claims included since its release in February 2025

Lab claims count from the day the card came out, independent runs from the model’s release. We draw one row each for the benchmark’s own leaderboard and what labs claim in their cards.

DeepSeek Google DeepMind OpenAI ByteDance Seed Qwen
Own leaderboard Not tested since Dec 2025
Gemini 3 Pro 73.8%
Lab claims 6 leaders

Sources: Official SuperGPQA leaderboard and labs’ model cards

Scores over time

All the scores we have, one dot each, so a model tested at several settings appears more than once.

Plot with other benchmarks
On the benchmark’s own leaderboard Reported by the lab in its model card Best so far
0% 20% 40% 60% 80% 100% 2024 2025 2026 Near the ceiling (85%) Qwen 3.6 Max (Preview) GPT-5.1 Seed-Thinking-v1.5 Claude 3.5 Sonnet (October 2024) Llama 3.1-405B Mixtral 8x22B Mixtral 8x7B 0% 20% 40% 60% 80% 100% 2024 2025 2026 Near the ceiling (85%) Qwen 3.6 Max (Preview) GPT-5.1

Sources: Official SuperGPQA leaderboard and labs’ model cards

4 labs put it in a launch post in the last six months

Launch posts from every lab that has listed it, one tick per post. Tall colored ticks include it, and short gray ones leave it out. Qwen accounts for 20 of the 36 releases that named it since 2024.

ByteDance Seed
Z.ai
Xiaomi Corp
Moonshot AI
Qwen
DeepSeek
OpenBMB (ModelBest)
DateLaunch postScoreAs the post writes it
Oct 7, 2026 MiniCPM-V-4.7-35B-A3B 61.2 SuperGPQA
Sep 10, 2026 DeepSeek V4.1 Flash 53.1 SuperGPQA (EM)
Jun 23, 2026 Seed2.1 Pro 70.8 SuperGPQA
Jun 1, 2026 Qwen3.7-Plus 71.4 SuperGPQA
May 20, 2026 Qwen3.7-Max 73.6 SuperGPQA
Apr 28, 2026 Seed2.0 Lite (0428) 69.6 SuperGPQA
Apr 24, 2026 DeepSeek-V4-Pro 53.9 SuperGPQA (EM)
Apr 22, 2026 Qwen3.6 27B 66 SuperGPQA
Apr 20, 2026 Qwen 3.6 Max (Preview) 73.9 SuperGPQA Graduate-Level Knowledge
Apr 15, 2026 Qwen 3.6 35B-A3B 64.7 SuperGPQA
Apr 2, 2026 Qwen 3.6 Plus 71.6 SuperGPQA
Mar 30, 2026 Qwen3.5-Omni-Plus 66.4 SuperGPQA
Mar 2, 2026 Qwen3.5-0.8B 21.3 SuperGPQA (Thinking)
Mar 2, 2026 Qwen3.5-2B 37.5 SuperGPQA (Thinking)
Mar 2, 2026 Qwen3.5-4B 52.9 SuperGPQA
Mar 2, 2026 Qwen3.5-9B 58.2 SuperGPQA
Feb 24, 2026 Qwen3.5-35B-A3B 63.4 SuperGPQA
Feb 24, 2026 Qwen3.5-27B 65.6 SuperGPQA
Feb 24, 2026 Qwen3.5-122B-A10B 67.1 SuperGPQA
Feb 16, 2026 Qwen3.5 397B-A17B 70.4 SuperGPQA
Feb 14, 2026 Seed2.0 Pro 68.7 SuperGPQA
Dec 18, 2025 Seed1.8 64.8 SuperGPQA
Sep 24, 2025 Qwen3-Max SuperGPQA
Sep 23, 2025 Qwen3-VL-235B-A22B 64.3 SuperGPQA
Sep 11, 2025 Qwen3-Next-80B-A3B 60.8 SuperGPQA
Aug 20, 2025 seed-oss-36b-instruct 55.7 SuperGPQA
Jul 25, 2025 Qwen3-235B-A22B-Thinking 64.9 SuperGPQA
Jul 11, 2025 Kimi K2 57.2 SuperGPQA, Acc
Apr 29, 2025 MiMo-7B-RL 40.5 SuperGPQA (Pass@1)
Apr 14, 2025 GLM-Z1-32B-0414 52.6 SuperGPQA
Apr 10, 2025 Seed-Thinking-v1.5 62.1 SuperGPQA

14 system cards report it, from 8 labs

Besides launch posts, we read the system cards, model cards and technical reports labs publish with a release. They report far more, including safety and frontier-risk tests.

DateDocumentScoreAs the document writes itSection
Oct 15, 2026 Nemotron 3.5 Super 120B-A12B Base technical report 47.6 SuperGPQA, CoT, Table 6 Capabilities
Oct 15, 2026 Nemotron 3.5 Lightning 30B-A3B Base technical report 33.2 SuperGPQA, CoT, Table 6 Capabilities
Sep 10, 2026 DeepSeek-V4.1-Flash Base model card 53.1 SuperGPQA, EM, 5-shot Capabilities
Jun 23, 2026 Seed2.1 Pro model card 70.8 SuperGPQA (Table 11) Capabilities
Jun 13, 2026 Ling-2.6-1T technical report 58.3 SuperGPQA, EM, CoT (Table 5) Capabilities
Apr 17, 2026 Qwen3.5-Omni-Plus technical report 66.4 SuperGPQA (Table 4) Capabilities
Mar 21, 2026 Seed1.8 model card 64.8 SuperGPQA (Table 1) Capabilities
Feb 28, 2026 Qwen3-Coder-Next technical report 57.5 SuperGPQA (Table 8) Capabilities
Feb 14, 2026 Seed2.0 Pro model card 68.7 SuperGPQA (Table 3) Capabilities
Jan 14, 2026 STEP3-VL-10B technical report 50.4 SuperGPQA, SeRe (Tables 2 and 3). PaCoRe 53.15 Capabilities
Nov 26, 2025 Qwen3-VL-235B-A22B technical report 64.3 SuperGPQA (Table 6) Capabilities
Jul 21, 2025 Kimi K2 technical report 57.2 SuperGPQA (Acc.) (Table 3) Capabilities
May 12, 2025 MiMo-7B-RL technical report 40.5 SuperGPQA (Pass@1) (Table 4) Capabilities
Apr 10, 2025 Seed-Thinking-v1.5 technical report 62.1 SuperGPQA (Table 2) Capabilities

One lab claim next to the leaderboard

Each score a lab reported for its own model, beside the same model's best score on the leaderboard. Gaps mostly come from different harnesses, effort levels and task subsets. Every benchmark.

Model Lab says Board says Gap
Kimi K2 Jul 2025 57.2% SuperGPQA, Acc 58.1% Leaderboard −0.9 pts

More in Knowledge

All 332 in Knowledge
  • MMLU-Pro In use Harder 10-option version of MMLU with reasoning-focused questions across 14 subjects. Intern-S2-397B 89.8%
  • MMLU In use 57-subject multiple-choice exam questions from elementary to professional level. Scored by accuracy. Seed1.8 92.3%
  • WideSearch In use 200 bilingual broad info-seeking tasks: agent collects many atomic facts into a table. Scored by success rate and row/item F1. Tencent Hy4 preview 83.9%
  • BBEH micro average In use Models solve harder replacements for each of the 23 BIG-Bench Hard tasks, 4,520 examples in total. The score is the share of all examples answered correctly. Gemma 4 31B IT 74.4%
  • HLE-Diamond In use Models answer 1,000 cleaned expert questions from Humanity's Last Exam, 500 on reasoning and 500 on knowledge, closed book. Graded on exact answers. Claude Fable 5.2 77.9%
  • HLE-Verified In use Verified and revised version of Humanity's Last Exam (1,811 items) with uncertain items removed. Gemini 3 Pro 48.2%

Data audit

All 11 checks pass. Every benchmark

Catalog record 2 pass
  • Pass Task access All tasks are public.
  • Pass Confidence Checked against a primary source.
Scores 4 pass 1 note
  • Pass Independent runs Scores from the official leaderboard (90 entries), launch posts (30 scores) and system cards (14 scores).
  • Pass Models scored 94 models we track, plus 31 other entries on its leaderboard.
  • Pass Newest score The newest is from October 2026, for Nemotron 3.5 Lightning 30B-A3B Base.
  • Note Frontier models None of the top 10 has a score. The strongest tested is GPT-5.2 Pro (#32).
  • Pass Headroom The best is 73.9%, under 85%, so it still separates models.
Leaderboard import 3 pass
  • Pass Importer We read the board on supergpqa.github.io: 70 rows for models we track and 31 other entries at the last read.
  • Pass Last read November 18th, 2026.
  • Pass Read worked The last read worked.
Consistency 2 pass
  • Pass Lab claims The one pair agrees within 10 points.
  • Pass Known issues No published audit found problems.

How we check