Benchmark Almanac

AIME 2025

Grok 4 Heavy 100% leads since July 2025
GPT-5.2 100% leads since December 2025
Grok 4 Heavy 100% leads since July 2025
In use

Models answer the 30 problems from the 2025 AIME I and II exams. Each answer is an integer from 0 to 999, scored by match.

The problems were published in February 2025, so models trained later may have seen them.

Released
February 2025
Kind
Competition problems with answers
Level
Competition
Scored
Checked automatically
Answers
Short answer
Tasks
All public
Strongest model tested
GPT-5.2 Pro #31 on the capability index
Last in a launch post
August 2026
Last model added
August 2026
Scores refreshed
November 12th, 2026
143
Models scored in lab cards
129
Launch posts list it, from 37 labs
81
System cards report it, from 29 labs

8 models share the top score of 100% among models released by August 2026

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

GPT-5.2 leads with 100% across all independent runs

Best score per model across MathArena’s leaderboard, Epoch AI’s own runs and any board Epoch AI copied. Each bar and row names its source.

Grok 4 Heavy leads with 100%, 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.

The newest model with a score here came out in August 2026, and nobody has reported a newer one. Among the 20 highest-ranked models on the capability index, these have no score yet:

143 models. Scroll sideways for the rest.

  1. 100.0 Grok 4 Heavy
  2. 100.0 GPT-5
  3. 100.0 Qwen3-Max
  4. 100.0 Claude Sonnet 4.5
  5. 100.0 Gemini 3 Pro
  6. 100.0 GPT-5.2
  7. 100.0 GPT-5.2 Pro
  8. 100.0 LongCat-Flash-Thinking-2601
  9. 99.8 Claude Opus 4.6
  10. 99.7 Gemini 3 Flash
  11. 99.2 Gemini 2.5 Deep Think
  12. 98.8 Grok 4
  13. 98.7 gpt-oss-20b
  14. 98.3 Seed2.0 Pro
  15. 97.9 gpt-oss-120b
  16. 97.3 Step 3.5 Flash
  17. 97.0 MAI-Thinking-1
  18. 96.3 Claude Haiku 4.5
  19. 96.3 Trinity Large Thinking
  20. 96.1 Kimi K2.5
  21. 96.0 DeepSeek-V3.2-Speciale
  22. 95.7 GLM-4.7
  23. 95.6 Claude Sonnet 4.6
  24. 94.5 Kimi K2 Thinking
  25. 94.3 Seed1.8
  26. 94.1 MiMo-V2-Flash
  27. 94.0 GPT-5.1
  28. 94.0 MiniMax-M2.7
  29. 93.9 GLM-4.6
  30. 93.4 Ring-1T
  31. 93.3 Grok 3
  32. 93.1 DeepSeek-V3.2
  33. 93.1 Intern-S1-Pro
  34. 92.9 EXAONE 4.5 33B
  35. 92.8 K-EXAONE
  36. 92.8 Claude Opus 4.5
  37. 92.7 o4-mini
  38. 92.4 Nemotron-Cascade 2 30B-A3B
  39. 92.3 Qwen3-235B-A22B-Thinking
  40. 92.3 Nova 2.0 Pro Preview
  41. 92.1 Nova 2 Omni Preview
  42. 92.0 Grok 4 Fast
  43. 91.6 glm-4.7-flash
  44. 91.1 GPT-5 mini
  45. 91.0 Nova 2 Lite
  46. 90.8 Grok-3 mini
  47. 90.6 LongCat-Flash-Thinking
  48. 90.2 Nemotron 3 Super
  49. 90.0 Claude Opus 4
  50. 90.0 Command A+
  51. 89.7 Qwen3-VL-235B-A22B
  52. 89.5 Cogito v2.1
  53. 89.3 DeepSeek-V3.2-Exp
  54. 89.2 Granite 4.2 30B
  55. 89.1 Nemotron 3 Nano 30B-A3B
  56. 89.1 ERNIE 5.0
  57. 88.9 o3
  58. 88.4 DeepSeek-V3.1
  59. 88.3 Sarvam 105B
  60. 88.0 Gemini 2.5 Pro
  61. 88.0 INTELLECT-3
  62. 87.8 Qwen3-Next-80B-A3B
  63. 87.7 STEP3-VL-10B
  64. 87.5 DeepSeek-R1
  65. 86.7 Gemini 2.5 Pro
  66. 86.7 Keye-VL-2.0-30B-A3B
  67. 86.3 MiniMax-M2.5
  68. 86.3 Mistral Medium 3.5
  69. 86.0 Seed1.6-Thinking
  70. 86.0 Intern-S1
  71. 85.3 EXAONE 4.0 32B
  72. 85.2 GPT-5 nano
  73. 85.0 Claude Sonnet 4
  74. 85.0 Ministral 3 14B
  75. 84.7 seed-oss-36b-instruct
  76. 84.3 Solar Open 100B
  77. 83.8 Mistral Small 4
  78. 83.5 Magistral Medium 1.2
  79. 83.3 Magistral Medium 1.0
  80. 83.1 Falcon H1R 7B
  81. 83.1 Qwen3-Coder-Next
  82. 83.0 MiniMax-M2.1
  83. 82.9 Step 3
  84. 82.7 Llama-3.3-Nemotron-Super-49B-v1.5
  85. 82.1 Nemotron 3 Nano Omni 30B-A3B
  86. 81.5 Qwen3-235B-A22B
  87. 81.2 GPT-5.5 Instant
  88. 80.3 Solar Pro 3
  89. 80.0 Sarvam 30B
  90. 78.5 Nemotron 3 Nano 4B
  91. 78.1 Hermes 4 405B
  92. 78.1 Olmo 3.1 32B Think
  93. 78.0 ERNIE-4.5-21B-A3B-Thinking
  94. 78.0 Gemini 2.5 Flash
  95. 78.0 Phi-4-Reasoning-plus
  96. 78.0 Claude Opus 4.1
  97. 78.0 MiniMax-M2
  98. 77.3 Magistral Small 1.2
  99. 76.9 MiniMax-M1-80k
  100. 76.8 Hunyuan-A13B-Instruct
  101. 75.6 InternVL3.5-241B-A28B
  102. 75.6 Gemini 2.5 Flash
  103. 74.0 Seed-Thinking-v1.5
  104. 72.9 Qwen3-32B
  105. 72.9 Baichuan-M2-32B
  106. 72.5 Llama-3.1-Nemotron-Ultra-253B-v1
  107. 72.5 Olmo 3 32B Think
  108. 72.1 nvidia-nemotron-nano-9b-v2
  109. 72.0 Gemini 2.5 Flash
  110. 71.3 Nemotron Nano 12B v2 VL
  111. 70.9 Qwen3-30B-A3B
  112. 70.4 Qwen3-14B
  113. 69.3 Hermes 4.3 36B
  114. 69.1 Solar Pro 2
  115. 68.2 Code World Model (CWM)
  116. 67.3 Qwen3-8B
  117. 65.8 EXAONE Deep 32B
  118. 65.6 Qwen3-4B
  119. 65.0 Qwen3-Omni-30B-A3B
  120. 64.9 INTELLECT-2
  121. 63.6 GLM-Z1-32B-0414
  122. 63.1 Phi-4-reasoning
  123. 63.1 Gemini 2.5 Flash-Lite
  124. 62.8 Magistral Small 1.0
  125. 62.0 Magistral Small 1.1
  126. 61.3 LongCat-Flash-Chat
  127. 61.0 Magistral Medium 1.1
  128. 60.0 Llama-3.3-Nemotron-Super-49B-v1
  129. 55.4 MiMo-7B-RL
  130. 51.9 LFM2.5-2.6B
  131. 50.0 Reka Flash 3
  132. 49.5 Kimi K2
  133. 48.2 Gemini 2.5 Flash-Lite
  134. 42.5 LFM2.5-8B-A1B
  135. 36.8 Qwen3-1.7B
  136. 35.1 ERNIE-4.5-300B-A47B
  137. 33.6 Phi-4-mini-flash-reasoning
  138. 31.7 LFM2.5-1.2B-Thinking
  139. 24.4 Trinity Large Preview
  140. 16.7 Falcon-H1-34B
  141. 16.0 Amazon Nova Premier
  142. 14.0 LFM2.5-1.2B-Instruct
  143. 11.6 Gemma 3n E4B
  1. 1 Grok 4 Heavy 100%
  2. 2 GPT-5 100%
  3. 3 Qwen3-Max 100%
  4. 4 Claude Sonnet 4.5 100%
  5. 5 Gemini 3 Pro 100%
  6. 6 GPT-5.2 100%
  7. 7 GPT-5.2 Pro 100%
  8. 8 LongCat-Flash-Thinking-2601 100%
  9. 9 Claude Opus 4.6 99.8%
  10. 10 Gemini 3 Flash 99.7%
  11. 11 Gemini 2.5 Deep Think 99.2%
  12. 12 Grok 4 98.8%
  13. 13 gpt-oss-20b 98.7%
  14. 14 Seed2.0 Pro 98.3%
  15. 15 gpt-oss-120b 97.9%
  16. 16 Step 3.5 Flash 97.3%
  17. 17 MAI-Thinking-1 97%
  18. 18 Claude Haiku 4.5 96.3%
  19. 19 Trinity Large Thinking 96.3%
  20. 20 Kimi K2.5 96.1%
  21. 21 DeepSeek-V3.2-Speciale 96%
  22. 22 GLM-4.7 95.7%
  23. 23 Claude Sonnet 4.6 95.6%
  24. 24 Kimi K2 Thinking 94.5%
  25. 25 Seed1.8 94.3%
  26. 26 MiMo-V2-Flash 94.1%
  27. 27 GPT-5.1 94%
  28. 28 MiniMax-M2.7 94%
  29. 29 GLM-4.6 93.9%
  30. 30 Ring-1T 93.4%
  31. 31 Grok 3 93.3%
  32. 32 DeepSeek-V3.2 93.1%
  33. 33 Intern-S1-Pro 93.1%
  34. 34 EXAONE 4.5 33B 92.9%
  35. 35 K-EXAONE 92.8%
  36. 36 Claude Opus 4.5 92.8%
  37. 37 o4-mini 92.7%
  38. 38 Nemotron-Cascade 2 30B-A3B 92.4%
  39. 39 Qwen3-235B-A22B-Thinking Jul 2025 92.3%
  40. 40 Nova 2.0 Pro Preview 92.3%
  41. 41 Nova 2 Omni Preview 92.1%
  42. 42 Grok 4 Fast 92%
  43. 43 glm-4.7-flash 91.6%
  44. 44 GPT-5 mini 91.1%
  45. 45 Nova 2 Lite 91%
  46. 46 Grok-3 mini 90.8%
  47. 47 LongCat-Flash-Thinking 90.6%
  48. 48 Nemotron 3 Super 90.2%
  49. 49 Claude Opus 4 90%
  50. 50 Command A+ 90%
  51. 51 Qwen3-VL-235B-A22B 89.7%
  52. 52 Cogito v2.1 89.5%
  53. 53 DeepSeek-V3.2-Exp 89.3%
  54. 54 Granite 4.2 30B 89.2%
  55. 55 Nemotron 3 Nano 30B-A3B 89.1%
  56. 56 ERNIE 5.0 89.1%
  57. 57 o3 88.9%
  58. 58 DeepSeek-V3.1 88.4%
  59. 59 Sarvam 105B 88.3%
  60. 60 Gemini 2.5 Pro Jun 2025 88%
  61. 61 INTELLECT-3 88%
  62. 62 Qwen3-Next-80B-A3B 87.8%
  63. 63 STEP3-VL-10B 87.7%
  64. 64 DeepSeek-R1 May 2025 87.5%
  65. 65 Gemini 2.5 Pro Mar 2025 86.7%
  66. 66 Keye-VL-2.0-30B-A3B 86.7%
  67. 67 MiniMax-M2.5 86.3%
  68. 68 Mistral Medium 3.5 86.3%
  69. 69 Seed1.6-Thinking 86%
  70. 70 Intern-S1 86%
  71. 71 EXAONE 4.0 32B 85.3%
  72. 72 GPT-5 nano 85.2%
  73. 73 Claude Sonnet 4 85%
  74. 74 Ministral 3 14B 85%
  75. 75 seed-oss-36b-instruct 84.7%
  76. 76 Solar Open 100B 84.3%
  77. 77 Mistral Small 4 83.8%
  78. 78 Magistral Medium 1.2 83.5%
  79. 79 Magistral Medium 1.0 83.3%
  80. 80 Falcon H1R 7B 83.1%
  81. 81 Qwen3-Coder-Next 83.1%
  82. 82 MiniMax-M2.1 83%
  83. 83 Step 3 82.9%
  84. 84 Llama-3.3-Nemotron-Super-49B-v1.5 82.7%
  85. 85 Nemotron 3 Nano Omni 30B-A3B 82.1%
  86. 86 Qwen3-235B-A22B 81.5%
  87. 87 GPT-5.5 Instant 81.2%
  88. 88 Solar Pro 3 80.3%
  89. 89 Sarvam 30B 80%
  90. 90 Nemotron 3 Nano 4B 78.5%
  91. 91 Hermes 4 405B 78.1%
  92. 92 Olmo 3.1 32B Think 78.1%
  93. 93 ERNIE-4.5-21B-A3B-Thinking 78%
  94. 94 Gemini 2.5 Flash Apr 2025 78%
  95. 95 Phi-4-Reasoning-plus 78%
  96. 96 Claude Opus 4.1 78%
  97. 97 MiniMax-M2 78%
  98. 98 Magistral Small 1.2 77.3%
  99. 99 MiniMax-M1-80k 76.9%
  100. 100 Hunyuan-A13B-Instruct 76.8%
  101. 101 InternVL3.5-241B-A28B 75.6%
  102. 102 Gemini 2.5 Flash Sep 2025 75.6%
  103. 103 Seed-Thinking-v1.5 74%
  104. 104 Qwen3-32B 72.9%
  105. 105 Baichuan-M2-32B 72.9%
  106. 106 Llama-3.1-Nemotron-Ultra-253B-v1 72.5%
  107. 107 Olmo 3 32B Think 72.5%
  108. 108 nvidia-nemotron-nano-9b-v2 72.1%
  109. 109 Gemini 2.5 Flash Jun 2025 72%
  110. 110 Nemotron Nano 12B v2 VL 71.3%
  111. 111 Qwen3-30B-A3B 70.9%
  112. 112 Qwen3-14B 70.4%
  113. 113 Hermes 4.3 36B 69.3%
  114. 114 Solar Pro 2 69.1%
  115. 115 Code World Model (CWM) 68.2%
  116. 116 Qwen3-8B 67.3%
  117. 117 EXAONE Deep 32B 65.8%
  118. 118 Qwen3-4B 65.6%
  119. 119 Qwen3-Omni-30B-A3B 65%
  120. 120 INTELLECT-2 64.9%
  121. 121 GLM-Z1-32B-0414 63.6%
  122. 122 Phi-4-reasoning 63.1%
  123. 123 Gemini 2.5 Flash-Lite Jun 2025 63.1%
  124. 124 Magistral Small 1.0 62.8%
  125. 125 Magistral Small 1.1 62%
  126. 126 LongCat-Flash-Chat 61.3%
  127. 127 Magistral Medium 1.1 61%
  128. 128 Llama-3.3-Nemotron-Super-49B-v1 60%
  129. 129 MiMo-7B-RL 55.4%
  130. 130 LFM2.5-2.6B 51.9%
  131. 131 Reka Flash 3 50%
  132. 132 Kimi K2 Jul 2025 49.5%
  133. 133 Gemini 2.5 Flash-Lite Sep 2025 48.2%
  134. 134 LFM2.5-8B-A1B 42.5%
  135. 135 Qwen3-1.7B 36.8%
  136. 136 ERNIE-4.5-300B-A47B 35.1%
  137. 137 Phi-4-mini-flash-reasoning 33.6%
  138. 138 LFM2.5-1.2B-Thinking 31.7%
  139. 139 Trinity Large Preview 24.4%
  140. 140 Falcon-H1-34B 16.7%
  141. 141 Amazon Nova Premier 16%
  142. 142 LFM2.5-1.2B-Instruct 14%
  143. 143 Gemma 3n E4B 11.6%

Source: Labs’ model cards

53 models. Scroll sideways for the rest.

  1. 100.0 GPT-5.2
  2. 98.3 Step 3.5 Flash
  3. 97.5 Gemini 3 Flash
  4. 96.7 GLM-5
  5. 95.8 DeepSeek-V3.2-Speciale
  6. 95.8 Kimi K2.5
  7. 95.0 GPT-5
  8. 95.0 Gemini 3 Pro
  9. 94.2 GPT-5.1
  10. 94.2 DeepSeek-V3.2
  11. 93.3 GLM-4.5
  12. 92.5 Grok 4
  13. 92.5 Kimi K2 Thinking
  14. 91.7 o4-mini
  15. 91.7 DeepSeek-V3.2-Exp
  16. 91.7 GLM-4.6
  17. 90.8 DeepSeek-V3.1
  18. 90.8 Grok 4 Fast
  19. 90.0 gpt-oss-120b
  20. 89.2 o3
  21. 89.2 DeepSeek-R1
  22. 89.2 gpt-oss-20b
  23. 89.2 Grok 4.1 Fast
  24. 88.3 Gemini 2.5 Pro
  25. 87.5 GPT-5 mini
  26. 86.7 o3-mini
  27. 86.7 Falcon H1R 7B
  28. 85.0 GPT-5 nano
  29. 84.2 Claude Sonnet 4.5
  30. 83.3 Gemini 2.5 Pro
  31. 83.3 GLM-4.5-Air
  32. 83.3 K2-Think
  33. 81.7 o1
  34. 81.7 Grok-3 mini
  35. 80.8 Qwen3-235B-A22B
  36. 77.5 QED-Nano
  37. 70.8 Gemini 2.5 Flash
  38. 70.0 DeepSeek-R1
  39. 70.0 Qwen3-30B-A3B
  40. 70.0 Claude Opus 4
  41. 65.8 QwQ-32B
  42. 60.0 DeepSeek-R1-Distill-Qwen-32B
  43. 55.0 DeepSeek-R1-Distill-Llama-70B
  44. 53.3 Gemini 2.0 Flash Thinking
  45. 50.0 DeepSeek-V3
  46. 49.2 DeepSeek-R1-Distill-Qwen-14B
  47. 49.2 Claude 3.7 Sonnet
  48. 33.3 QwQ-32B-Preview
  49. 27.5 Gemini 2.0 Flash
  50. 27.5 Gemini 2.0 Pro
  51. 20.0 DeepSeek-R1-Distill-Qwen-1.5B
  52. 11.7 GPT-4o
  53. 3.33 Claude 3.5 Sonnet
  1. 1 GPT-5.2 high 100%
  2. 2 Step 3.5 Flash 98.3%
  3. 3 Gemini 3 Flash high 97.5%
  4. 4 GLM-5 96.7%
  5. 5 DeepSeek-V3.2-Speciale 95.8%
  6. 6 Kimi K2.5 high 95.8%
  7. 7 GPT-5 high 95%
  8. 8 Gemini 3 Pro 95%
  9. 9 GPT-5.1 high 94.2%
  10. 10 DeepSeek-V3.2 thinking 94.2%
  11. 11 GLM-4.5 93.3%
  12. 12 Grok 4 92.5%
  13. 13 Kimi K2 Thinking 92.5%
  14. 14 o4-mini high 91.7%
  15. 15 DeepSeek-V3.2-Exp thinking 91.7%
  16. 16 GLM-4.6 91.7%
  17. 17 DeepSeek-V3.1 thinking 90.8%
  18. 18 Grok 4 Fast 90.8%
  19. 19 gpt-oss-120b high 90%
  20. 20 o3 high 89.2%
  21. 21 DeepSeek-R1 May 2025 89.2%
  22. 22 gpt-oss-20b high 89.2%
  23. 23 Grok 4.1 Fast reasoning 89.2%
  24. 24 Gemini 2.5 Pro Mar 2025 88.3%
  25. 25 GPT-5 mini high 87.5%
  26. 26 o3-mini high 86.7%
  27. 27 Falcon H1R 7B 86.7%
  28. 28 GPT-5 nano high 85%
  29. 29 Claude Sonnet 4.5 thinking 84.2%
  30. 30 Gemini 2.5 Pro May 2025 83.3%
  31. 31 GLM-4.5-Air 83.3%
  32. 32 K2-Think 83.3%
  33. 33 o1 medium 81.7%
  34. 34 Grok-3 mini high 81.7%
  35. 35 Qwen3-235B-A22B 80.8%
  36. 36 QED-Nano 77.5%
  37. 37 Gemini 2.5 Flash Sep 2025, thinking 70.8%
  38. 38 DeepSeek-R1 70%
  39. 39 Qwen3-30B-A3B 70%
  40. 40 Claude Opus 4 thinking 70%
  41. 41 QwQ-32B 65.8%
  42. 42 DeepSeek-R1-Distill-Qwen-32B 60%
  43. 43 DeepSeek-R1-Distill-Llama-70B 55%
  44. 44 Gemini 2.0 Flash Thinking Jan 2025 53.3%
  45. 45 DeepSeek-V3 Mar 2025 50%
  46. 46 DeepSeek-R1-Distill-Qwen-14B 49.2%
  47. 47 Claude 3.7 Sonnet thinking 49.2%
  48. 48 QwQ-32B-Preview 33.3%
  49. 49 Gemini 2.0 Flash Feb 2025 27.5%
  50. 50 Gemini 2.0 Pro 27.5%
  51. 51 DeepSeek-R1-Distill-Qwen-1.5B 20%
  52. 52 GPT-4o Mar 2025 11.7%
  53. 53 Claude 3.5 Sonnet 3.33%

Source: MathArena’s leaderboard

166 models. Scroll sideways for the rest.

  1. 100.0 Grok 4 Heavy
  2. 100.0 GPT-5
  3. 100.0 Qwen3-Max
  4. 100.0 Claude Sonnet 4.5
  5. 100.0 Gemini 3 Pro
  6. 100.0 GPT-5.2
  7. 100.0 GPT-5.2 Pro
  8. 100.0 LongCat-Flash-Thinking-2601
  9. 99.8 Claude Opus 4.6
  10. 99.7 Gemini 3 Flash
  11. 99.2 Gemini 2.5 Deep Think
  12. 98.8 Grok 4
  13. 98.7 gpt-oss-20b
  14. 98.3 Step 3.5 Flash
  15. 98.3 Seed2.0 Pro
  16. 97.9 gpt-oss-120b
  17. 97.0 MAI-Thinking-1
  18. 96.7 GLM-5
  19. 96.3 Claude Haiku 4.5
  20. 96.3 Trinity Large Thinking
  21. 96.1 Kimi K2.5
  22. 96.0 DeepSeek-V3.2-Speciale
  23. 95.7 GLM-4.7
  24. 95.6 Claude Sonnet 4.6
  25. 94.5 Kimi K2 Thinking
  26. 94.3 Seed1.8
  27. 94.2 GPT-5.1
  28. 94.2 DeepSeek-V3.2
  29. 94.1 MiMo-V2-Flash
  30. 94.0 MiniMax-M2.7
  31. 93.9 GLM-4.6
  32. 93.4 Ring-1T
  33. 93.3 GLM-4.5
  34. 93.3 Grok 3
  35. 93.1 Intern-S1-Pro
  36. 92.9 EXAONE 4.5 33B
  37. 92.8 K-EXAONE
  38. 92.8 Claude Opus 4.5
  39. 92.7 o4-mini
  40. 92.4 Nemotron-Cascade 2 30B-A3B
  41. 92.3 Qwen3-235B-A22B-Thinking
  42. 92.3 Nova 2.0 Pro Preview
  43. 92.1 Nova 2 Omni Preview
  44. 92.0 Grok 4 Fast
  45. 91.7 DeepSeek-V3.2-Exp
  46. 91.6 glm-4.7-flash
  47. 91.1 GPT-5 mini
  48. 91.0 Nova 2 Lite
  49. 90.8 DeepSeek-V3.1
  50. 90.8 Grok-3 mini
  51. 90.6 LongCat-Flash-Thinking
  52. 90.2 Nemotron 3 Super
  53. 90.0 Claude Opus 4
  54. 90.0 Command A+
  55. 89.7 Qwen3-VL-235B-A22B
  56. 89.5 Cogito v2.1
  57. 89.2 o3
  58. 89.2 DeepSeek-R1
  59. 89.2 Grok 4.1 Fast
  60. 89.2 Granite 4.2 30B
  61. 89.1 Nemotron 3 Nano 30B-A3B
  62. 89.1 ERNIE 5.0
  63. 88.3 Gemini 2.5 Pro
  64. 88.3 Sarvam 105B
  65. 88.0 Gemini 2.5 Pro
  66. 88.0 INTELLECT-3
  67. 87.8 Qwen3-Next-80B-A3B
  68. 87.7 STEP3-VL-10B
  69. 86.7 Keye-VL-2.0-30B-A3B
  70. 86.7 o3-mini
  71. 86.7 Falcon H1R 7B
  72. 86.3 MiniMax-M2.5
  73. 86.3 Mistral Medium 3.5
  74. 86.0 Seed1.6-Thinking
  75. 86.0 Intern-S1
  76. 85.3 EXAONE 4.0 32B
  77. 85.2 GPT-5 nano
  78. 85.0 Claude Sonnet 4
  79. 85.0 Ministral 3 14B
  80. 84.7 seed-oss-36b-instruct
  81. 84.3 Solar Open 100B
  82. 83.8 Mistral Small 4
  83. 83.5 Magistral Medium 1.2
  84. 83.3 Gemini 2.5 Pro
  85. 83.3 GLM-4.5-Air
  86. 83.3 K2-Think
  87. 83.3 Magistral Medium 1.0
  88. 83.1 Qwen3-Coder-Next
  89. 83.0 MiniMax-M2.1
  90. 82.9 Step 3
  91. 82.7 Llama-3.3-Nemotron-Super-49B-v1.5
  92. 82.1 Nemotron 3 Nano Omni 30B-A3B
  93. 81.7 o1
  94. 81.5 Qwen3-235B-A22B
  95. 81.2 GPT-5.5 Instant
  96. 80.3 Solar Pro 3
  97. 80.0 Sarvam 30B
  98. 78.5 Nemotron 3 Nano 4B
  99. 78.1 Hermes 4 405B
  100. 78.1 Olmo 3.1 32B Think
  101. 78.0 ERNIE-4.5-21B-A3B-Thinking
  102. 78.0 Gemini 2.5 Flash
  103. 78.0 Phi-4-Reasoning-plus
  104. 78.0 Claude Opus 4.1
  105. 78.0 MiniMax-M2
  106. 77.5 QED-Nano
  107. 77.3 Magistral Small 1.2
  108. 76.9 MiniMax-M1-80k
  109. 76.8 Hunyuan-A13B-Instruct
  110. 75.6 InternVL3.5-241B-A28B
  111. 75.6 Gemini 2.5 Flash
  112. 74.0 Seed-Thinking-v1.5
  113. 72.9 Qwen3-32B
  114. 72.9 Baichuan-M2-32B
  115. 72.5 Llama-3.1-Nemotron-Ultra-253B-v1
  116. 72.5 Olmo 3 32B Think
  117. 72.1 nvidia-nemotron-nano-9b-v2
  118. 72.0 Gemini 2.5 Flash
  119. 71.3 Nemotron Nano 12B v2 VL
  120. 70.9 Qwen3-30B-A3B
  121. 70.4 Qwen3-14B
  122. 70.0 DeepSeek-R1
  123. 69.3 Hermes 4.3 36B
  124. 69.1 Solar Pro 2
  125. 68.2 Code World Model (CWM)
  126. 67.3 Qwen3-8B
  127. 65.8 QwQ-32B
  128. 65.8 EXAONE Deep 32B
  129. 65.6 Qwen3-4B
  130. 65.0 Qwen3-Omni-30B-A3B
  131. 64.9 INTELLECT-2
  132. 63.6 GLM-Z1-32B-0414
  133. 63.1 Phi-4-reasoning
  134. 63.1 Gemini 2.5 Flash-Lite
  135. 62.8 Magistral Small 1.0
  136. 62.0 Magistral Small 1.1
  137. 61.3 LongCat-Flash-Chat
  138. 61.0 Magistral Medium 1.1
  139. 60.0 DeepSeek-R1-Distill-Qwen-32B
  140. 60.0 Llama-3.3-Nemotron-Super-49B-v1
  141. 55.4 MiMo-7B-RL
  142. 55.0 DeepSeek-R1-Distill-Llama-70B
  143. 53.3 Gemini 2.0 Flash Thinking
  144. 51.9 LFM2.5-2.6B
  145. 50.0 Reka Flash 3
  146. 50.0 DeepSeek-V3
  147. 49.5 Kimi K2
  148. 49.2 DeepSeek-R1-Distill-Qwen-14B
  149. 49.2 Claude 3.7 Sonnet
  150. 48.2 Gemini 2.5 Flash-Lite
  151. 42.5 LFM2.5-8B-A1B
  152. 36.8 Qwen3-1.7B
  153. 35.1 ERNIE-4.5-300B-A47B
  154. 33.6 Phi-4-mini-flash-reasoning
  155. 33.3 QwQ-32B-Preview
  156. 31.7 LFM2.5-1.2B-Thinking
  157. 27.5 Gemini 2.0 Flash
  158. 27.5 Gemini 2.0 Pro
  159. 24.4 Trinity Large Preview
  160. 20.0 DeepSeek-R1-Distill-Qwen-1.5B
  161. 16.7 Falcon-H1-34B
  162. 16.0 Amazon Nova Premier
  163. 14.0 LFM2.5-1.2B-Instruct
  164. 11.7 GPT-4o
  165. 11.6 Gemma 3n E4B
  166. 3.33 Claude 3.5 Sonnet
  1. 1 Grok 4 Heavy 100%
  2. 2 GPT-5 100%
  3. 3 Qwen3-Max 100%
  4. 4 Claude Sonnet 4.5 100%
  5. 5 Gemini 3 Pro 100%
  6. 6 GPT-5.2 high 100%
  7. 7 GPT-5.2 Pro 100%
  8. 8 LongCat-Flash-Thinking-2601 100%
  9. 9 Claude Opus 4.6 99.8%
  10. 10 Gemini 3 Flash 99.7%
  11. 11 Gemini 2.5 Deep Think 99.2%
  12. 12 Grok 4 98.8%
  13. 13 gpt-oss-20b 98.7%
  14. 14 Step 3.5 Flash 98.3%
  15. 15 Seed2.0 Pro 98.3%
  16. 16 gpt-oss-120b 97.9%
  17. 17 MAI-Thinking-1 97%
  18. 18 GLM-5 96.7%
  19. 19 Claude Haiku 4.5 96.3%
  20. 20 Trinity Large Thinking 96.3%
  21. 21 Kimi K2.5 96.1%
  22. 22 DeepSeek-V3.2-Speciale 96%
  23. 23 GLM-4.7 95.7%
  24. 24 Claude Sonnet 4.6 95.6%
  25. 25 Kimi K2 Thinking 94.5%
  26. 26 Seed1.8 94.3%
  27. 27 GPT-5.1 high 94.2%
  28. 28 DeepSeek-V3.2 thinking 94.2%
  29. 29 MiMo-V2-Flash 94.1%
  30. 30 MiniMax-M2.7 94%
  31. 31 GLM-4.6 93.9%
  32. 32 Ring-1T 93.4%
  33. 33 GLM-4.5 93.3%
  34. 34 Grok 3 93.3%
  35. 35 Intern-S1-Pro 93.1%
  36. 36 EXAONE 4.5 33B 92.9%
  37. 37 K-EXAONE 92.8%
  38. 38 Claude Opus 4.5 92.8%
  39. 39 o4-mini 92.7%
  40. 40 Nemotron-Cascade 2 30B-A3B 92.4%
  41. 41 Qwen3-235B-A22B-Thinking Jul 2025 92.3%
  42. 42 Nova 2.0 Pro Preview 92.3%
  43. 43 Nova 2 Omni Preview 92.1%
  44. 44 Grok 4 Fast 92%
  45. 45 DeepSeek-V3.2-Exp thinking 91.7%
  46. 46 glm-4.7-flash 91.6%
  47. 47 GPT-5 mini 91.1%
  48. 48 Nova 2 Lite 91%
  49. 49 DeepSeek-V3.1 thinking 90.8%
  50. 50 Grok-3 mini 90.8%
  51. 51 LongCat-Flash-Thinking 90.6%
  52. 52 Nemotron 3 Super 90.2%
  53. 53 Claude Opus 4 90%
  54. 54 Command A+ 90%
  55. 55 Qwen3-VL-235B-A22B 89.7%
  56. 56 Cogito v2.1 89.5%
  57. 57 o3 high 89.2%
  58. 58 DeepSeek-R1 May 2025 89.2%
  59. 59 Grok 4.1 Fast reasoning 89.2%
  60. 60 Granite 4.2 30B 89.2%
  61. 61 Nemotron 3 Nano 30B-A3B 89.1%
  62. 62 ERNIE 5.0 89.1%
  63. 63 Gemini 2.5 Pro Mar 2025 88.3%
  64. 64 Sarvam 105B 88.3%
  65. 65 Gemini 2.5 Pro Jun 2025 88%
  66. 66 INTELLECT-3 88%
  67. 67 Qwen3-Next-80B-A3B 87.8%
  68. 68 STEP3-VL-10B 87.7%
  69. 69 Keye-VL-2.0-30B-A3B 86.7%
  70. 70 o3-mini high 86.7%
  71. 71 Falcon H1R 7B 86.7%
  72. 72 MiniMax-M2.5 86.3%
  73. 73 Mistral Medium 3.5 86.3%
  74. 74 Seed1.6-Thinking 86%
  75. 75 Intern-S1 86%
  76. 76 EXAONE 4.0 32B 85.3%
  77. 77 GPT-5 nano 85.2%
  78. 78 Claude Sonnet 4 85%
  79. 79 Ministral 3 14B 85%
  80. 80 seed-oss-36b-instruct 84.7%
  81. 81 Solar Open 100B 84.3%
  82. 82 Mistral Small 4 83.8%
  83. 83 Magistral Medium 1.2 83.5%
  84. 84 Gemini 2.5 Pro May 2025 83.3%
  85. 85 GLM-4.5-Air 83.3%
  86. 86 K2-Think 83.3%
  87. 87 Magistral Medium 1.0 83.3%
  88. 88 Qwen3-Coder-Next 83.1%
  89. 89 MiniMax-M2.1 83%
  90. 90 Step 3 82.9%
  91. 91 Llama-3.3-Nemotron-Super-49B-v1.5 82.7%
  92. 92 Nemotron 3 Nano Omni 30B-A3B 82.1%
  93. 93 o1 medium 81.7%
  94. 94 Qwen3-235B-A22B 81.5%
  95. 95 GPT-5.5 Instant 81.2%
  96. 96 Solar Pro 3 80.3%
  97. 97 Sarvam 30B 80%
  98. 98 Nemotron 3 Nano 4B 78.5%
  99. 99 Hermes 4 405B 78.1%
  100. 100 Olmo 3.1 32B Think 78.1%
  101. 101 ERNIE-4.5-21B-A3B-Thinking 78%
  102. 102 Gemini 2.5 Flash Apr 2025 78%
  103. 103 Phi-4-Reasoning-plus 78%
  104. 104 Claude Opus 4.1 78%
  105. 105 MiniMax-M2 78%
  106. 106 QED-Nano 77.5%
  107. 107 Magistral Small 1.2 77.3%
  108. 108 MiniMax-M1-80k 76.9%
  109. 109 Hunyuan-A13B-Instruct 76.8%
  110. 110 InternVL3.5-241B-A28B 75.6%
  111. 111 Gemini 2.5 Flash Sep 2025 75.6%
  112. 112 Seed-Thinking-v1.5 74%
  113. 113 Qwen3-32B 72.9%
  114. 114 Baichuan-M2-32B 72.9%
  115. 115 Llama-3.1-Nemotron-Ultra-253B-v1 72.5%
  116. 116 Olmo 3 32B Think 72.5%
  117. 117 nvidia-nemotron-nano-9b-v2 72.1%
  118. 118 Gemini 2.5 Flash Jun 2025 72%
  119. 119 Nemotron Nano 12B v2 VL 71.3%
  120. 120 Qwen3-30B-A3B 70.9%
  121. 121 Qwen3-14B 70.4%
  122. 122 DeepSeek-R1 70%
  123. 123 Hermes 4.3 36B 69.3%
  124. 124 Solar Pro 2 69.1%
  125. 125 Code World Model (CWM) 68.2%
  126. 126 Qwen3-8B 67.3%
  127. 127 QwQ-32B 65.8%
  128. 128 EXAONE Deep 32B 65.8%
  129. 129 Qwen3-4B 65.6%
  130. 130 Qwen3-Omni-30B-A3B 65%
  131. 131 INTELLECT-2 64.9%
  132. 132 GLM-Z1-32B-0414 63.6%
  133. 133 Phi-4-reasoning 63.1%
  134. 134 Gemini 2.5 Flash-Lite Jun 2025 63.1%
  135. 135 Magistral Small 1.0 62.8%
  136. 136 Magistral Small 1.1 62%
  137. 137 LongCat-Flash-Chat 61.3%
  138. 138 Magistral Medium 1.1 61%
  139. 139 DeepSeek-R1-Distill-Qwen-32B 60%
  140. 140 Llama-3.3-Nemotron-Super-49B-v1 60%
  141. 141 MiMo-7B-RL 55.4%
  142. 142 DeepSeek-R1-Distill-Llama-70B 55%
  143. 143 Gemini 2.0 Flash Thinking Jan 2025 53.3%
  144. 144 LFM2.5-2.6B 51.9%
  145. 145 Reka Flash 3 50%
  146. 146 DeepSeek-V3 Mar 2025 50%
  147. 147 Kimi K2 Jul 2025 49.5%
  148. 148 DeepSeek-R1-Distill-Qwen-14B 49.2%
  149. 149 Claude 3.7 Sonnet thinking 49.2%
  150. 150 Gemini 2.5 Flash-Lite Sep 2025 48.2%
  151. 151 LFM2.5-8B-A1B 42.5%
  152. 152 Qwen3-1.7B 36.8%
  153. 153 ERNIE-4.5-300B-A47B 35.1%
  154. 154 Phi-4-mini-flash-reasoning 33.6%
  155. 155 QwQ-32B-Preview 33.3%
  156. 156 LFM2.5-1.2B-Thinking 31.7%
  157. 157 Gemini 2.0 Flash Feb 2025 27.5%
  158. 158 Gemini 2.0 Pro 27.5%
  159. 159 Trinity Large Preview 24.4%
  160. 160 DeepSeek-R1-Distill-Qwen-1.5B 20%
  161. 161 Falcon-H1-34B 16.7%
  162. 162 Amazon Nova Premier 16%
  163. 163 LFM2.5-1.2B-Instruct 14%
  164. 164 GPT-4o Mar 2025 11.7%
  165. 165 Gemma 3n E4B 11.6%
  166. 166 Claude 3.5 Sonnet 3.33%

Sources: Labs’ model cards and MathArena’s leaderboard

#ModelScoreSettingDateSource
1 Grok 4 Heavy 100% Jul 2025 Lab model card
2 GPT-5 100% Aug 2025 Lab model card
3 Qwen3-Max 100% Sep 2025 Lab model card
4 Claude Sonnet 4.5 100% Sep 2025 Lab model card
5 Gemini 3 Pro 100% Nov 2025 Lab model card
6 GPT-5.2 100% Dec 2025 Lab model card
7 GPT-5.2 Pro 100% Dec 2025 Lab model card
8 LongCat-Flash-Thinking-2601 100% Jan 2026 Lab model card
9 Claude Opus 4.6 99.8% Feb 2026 Lab model card
10 Gemini 3 Flash 99.7% Dec 2025 Lab model card
11 Gemini 2.5 Deep Think 99.2% Aug 2025 Lab model card
12 Grok 4 98.8% Jul 2025 Lab model card
13 gpt-oss-20b 98.7% Aug 2025 Lab model card
14 Seed2.0 Pro 98.3% Feb 2026 Lab model card
15 gpt-oss-120b 97.9% Aug 2025 Lab model card
16 Step 3.5 Flash 97.3% Feb 2026 Lab model card
17 MAI-Thinking-1 97% Jun 2026 Lab model card
18 Claude Haiku 4.5 96.3% Oct 2025 Lab model card
19 Trinity Large Thinking 96.3% Apr 2026 Lab model card
20 Kimi K2.5 96.1% Jan 2026 Lab model card
21 DeepSeek-V3.2-Speciale 96% Dec 2025 Lab model card
22 GLM-4.7 95.7% Dec 2025 Lab model card
23 Claude Sonnet 4.6 95.6% Feb 2026 Lab model card
24 Kimi K2 Thinking 94.5% Nov 2025 Lab model card
25 Seed1.8 94.3% Dec 2025 Lab model card
26 MiMo-V2-Flash 94.1% Dec 2025 Lab model card
27 GPT-5.1 94% Nov 2025 Lab model card
28 MiniMax-M2.7 94% May 2026 Lab model card
29 GLM-4.6 93.9% Sep 2025 Lab model card
30 Ring-1T 93.4% Oct 2025 Lab model card
31 Grok 3 93.3% Feb 2025 Lab model card
32 DeepSeek-V3.2 93.1% Dec 2025 Lab model card
33 Intern-S1-Pro 93.1% Feb 2026 Lab model card
34 EXAONE 4.5 33B 92.9% Apr 2026 Lab model card
35 K-EXAONE 92.8% Dec 2025 Lab model card
36 Claude Opus 4.5 92.8% Nov 2025 Lab model card
37 o4-mini 92.7% Apr 2025 Lab model card
38 Nemotron-Cascade 2 30B-A3B 92.4% Mar 2026 Lab model card
39 Qwen3-235B-A22B-Thinking Jul 2025 92.3% Jul 2025 Lab model card
40 Nova 2.0 Pro Preview 92.3% Dec 2025 Lab model card
41 Nova 2 Omni Preview 92.1% Dec 2025 Lab model card
42 Grok 4 Fast 92% Sep 2025 Lab model card
43 glm-4.7-flash 91.6% Jan 2026 Lab model card
44 GPT-5 mini 91.1% Aug 2025 Lab model card
45 Nova 2 Lite 91% Dec 2025 Lab model card
46 Grok-3 mini 90.8% Feb 2025 Lab model card
47 LongCat-Flash-Thinking 90.6% Sep 2025 Lab model card
48 Nemotron 3 Super 90.2% Mar 2026 Lab model card
49 Claude Opus 4 90% May 2025 Lab model card
50 Command A+ 90% May 2026 Lab model card
51 Qwen3-VL-235B-A22B 89.7% Sep 2025 Lab model card
52 Cogito v2.1 89.5% Nov 2025 Lab model card
53 DeepSeek-V3.2-Exp 89.3% Sep 2025 Lab model card
54 Granite 4.2 30B 89.2% Aug 2026 Lab model card
55 Nemotron 3 Nano 30B-A3B 89.1% Dec 2025 Lab model card
56 ERNIE 5.0 89.1% Feb 2026 Lab model card
57 o3 88.9% Apr 2025 Lab model card
58 DeepSeek-V3.1 88.4% Aug 2025 Lab model card
59 Sarvam 105B 88.3% Mar 2026 Lab model card
60 Gemini 2.5 Pro Jun 2025 88% Jun 2025 Lab model card
61 INTELLECT-3 88% Nov 2025 Lab model card
62 Qwen3-Next-80B-A3B 87.8% Sep 2025 Lab model card
63 STEP3-VL-10B 87.7% Jan 2026 Lab model card
64 DeepSeek-R1 May 2025 87.5% May 2025 Lab model card
65 Gemini 2.5 Pro Mar 2025 86.7% Mar 2025 Lab model card
66 Keye-VL-2.0-30B-A3B 86.7% May 2026 Lab model card
67 MiniMax-M2.5 86.3% Feb 2026 Lab model card
68 Mistral Medium 3.5 86.3% May 2026 Lab model card
69 Seed1.6-Thinking 86% Jun 2025 Lab model card
70 Intern-S1 86% Jul 2025 Lab model card
71 EXAONE 4.0 32B 85.3% Jul 2025 Lab model card
72 GPT-5 nano 85.2% Aug 2025 Lab model card
73 Claude Sonnet 4 85% May 2025 Lab model card
74 Ministral 3 14B 85% Dec 2025 Lab model card
75 seed-oss-36b-instruct 84.7% Aug 2025 Lab model card
76 Solar Open 100B 84.3% Dec 2025 Lab model card
77 Mistral Small 4 83.8% Mar 2026 Lab model card
78 Magistral Medium 1.2 83.5% Sep 2025 Lab model card
79 Magistral Medium 1.0 83.3% Jun 2025 Lab model card
80 Falcon H1R 7B 83.1% Jan 2026 Lab model card
81 Qwen3-Coder-Next 83.1% Feb 2026 Lab model card
82 MiniMax-M2.1 83% Dec 2025 Lab model card
83 Step 3 82.9% Jul 2025 Lab model card
84 Llama-3.3-Nemotron-Super-49B-v1.5 82.7% Jul 2025 Lab model card
85 Nemotron 3 Nano Omni 30B-A3B 82.1% Apr 2026 Lab model card
86 Qwen3-235B-A22B 81.5% Apr 2025 Lab model card
87 GPT-5.5 Instant 81.2% May 2026 Lab model card
88 Solar Pro 3 80.3% Jan 2026 Lab model card
89 Sarvam 30B 80% Mar 2026 Lab model card
90 Nemotron 3 Nano 4B 78.5% Mar 2026 Lab model card
91 Hermes 4 405B 78.1% Aug 2025 Lab model card
92 Olmo 3.1 32B Think 78.1% Dec 2025 Lab model card
93 ERNIE-4.5-21B-A3B-Thinking 78% Sep 2025 Lab model card
94 Gemini 2.5 Flash Apr 2025 78% Apr 2025 Lab model card
95 Phi-4-Reasoning-plus 78% Apr 2025 Lab model card
96 Claude Opus 4.1 78% Aug 2025 Lab model card
97 MiniMax-M2 78% Oct 2025 Lab model card
98 Magistral Small 1.2 77.3% Sep 2025 Lab model card
99 MiniMax-M1-80k 76.9% Jun 2025 Lab model card
100 Hunyuan-A13B-Instruct 76.8% Jun 2025 Lab model card
101 InternVL3.5-241B-A28B 75.6% Aug 2025 Lab model card
102 Gemini 2.5 Flash Sep 2025 75.6% Sep 2025 Lab model card
103 Seed-Thinking-v1.5 74% Apr 2025 Lab model card
104 Qwen3-32B 72.9% May 2025 Lab model card
105 Baichuan-M2-32B 72.9% Aug 2025 Lab model card
106 Llama-3.1-Nemotron-Ultra-253B-v1 72.5% Apr 2025 Lab model card
107 Olmo 3 32B Think 72.5% Nov 2025 Lab model card
108 nvidia-nemotron-nano-9b-v2 72.1% Aug 2025 Lab model card
109 Gemini 2.5 Flash Jun 2025 72% Jun 2025 Lab model card
110 Nemotron Nano 12B v2 VL 71.3% Nov 2025 Lab model card
111 Qwen3-30B-A3B 70.9% May 2025 Lab model card
112 Qwen3-14B 70.4% May 2025 Lab model card
113 Hermes 4.3 36B 69.3% Dec 2025 Lab model card
114 Solar Pro 2 69.1% Jul 2025 Lab model card
115 Code World Model (CWM) 68.2% Sep 2025 Lab model card
116 Qwen3-8B 67.3% May 2025 Lab model card
117 EXAONE Deep 32B 65.8% Mar 2025 Lab model card
118 Qwen3-4B 65.6% May 2025 Lab model card
119 Qwen3-Omni-30B-A3B 65% Sep 2025 Lab model card
120 INTELLECT-2 64.9% May 2025 Lab model card
121 GLM-Z1-32B-0414 63.6% Apr 2025 Lab model card
122 Phi-4-reasoning 63.1% Apr 2025 Lab model card
123 Gemini 2.5 Flash-Lite Jun 2025 63.1% Jun 2025 Lab model card
124 Magistral Small 1.0 62.8% Jun 2025 Lab model card
125 Magistral Small 1.1 62% Jul 2025 Lab model card
126 LongCat-Flash-Chat 61.3% Aug 2025 Lab model card
127 Magistral Medium 1.1 61% Jul 2025 Lab model card
128 Llama-3.3-Nemotron-Super-49B-v1 60% May 2025 Lab model card
129 MiMo-7B-RL 55.4% Apr 2025 Lab model card
130 LFM2.5-2.6B 51.9% Aug 2026 Lab model card
131 Reka Flash 3 50% Mar 2025 Lab model card
132 Kimi K2 Jul 2025 49.5% Jul 2025 Lab model card
133 Gemini 2.5 Flash-Lite Sep 2025 48.2% Sep 2025 Lab model card
134 LFM2.5-8B-A1B 42.5% May 2026 Lab model card
135 Qwen3-1.7B 36.8% May 2025 Lab model card
136 ERNIE-4.5-300B-A47B 35.1% Jun 2025 Lab model card
137 Phi-4-mini-flash-reasoning 33.6% Jul 2025 Lab model card
138 LFM2.5-1.2B-Thinking 31.7% Jan 2026 Lab model card
139 Trinity Large Preview 24.4% Feb 2026 Lab model card
140 Falcon-H1-34B 16.7% May 2025 Lab model card
141 Amazon Nova Premier 16% Apr 2025 Lab model card
142 LFM2.5-1.2B-Instruct 14% Jan 2026 Lab model card
143 Gemma 3n E4B 11.6% Jun 2025 Lab model card
#ModelScoreSettingDateSource
1 GPT-5.2 100% high Dec 2025 Official leaderboard
2 Step 3.5 Flash 98.3% Feb 2026 Official leaderboard
3 Gemini 3 Flash 97.5% high Dec 2025 Official leaderboard
4 GLM-5 96.7% Feb 2026 Official leaderboard
5 DeepSeek-V3.2-Speciale 95.8% Dec 2025 Official leaderboard
6 Kimi K2.5 95.8% high Jan 2026 Official leaderboard
7 GPT-5 95% high Aug 2025 Official leaderboard
8 Gemini 3 Pro 95% Nov 2025 Official leaderboard
9 GPT-5.1 94.2% high Nov 2025 Official leaderboard
10 DeepSeek-V3.2 94.2% thinking Dec 2025 Official leaderboard
11 GLM-4.5 93.3% Aug 2025 Official leaderboard
12 Grok 4 92.5% Jul 2025 Official leaderboard
13 Kimi K2 Thinking 92.5% Nov 2025 Official leaderboard
14 o4-mini 91.7% high Apr 2025 Official leaderboard
15 DeepSeek-V3.2-Exp 91.7% thinking Sep 2025 Official leaderboard
16 GLM-4.6 91.7% Sep 2025 Official leaderboard
17 DeepSeek-V3.1 90.8% thinking Aug 2025 Official leaderboard
18 Grok 4 Fast 90.8% Sep 2025 Official leaderboard
19 gpt-oss-120b 90% high Aug 2025 Official leaderboard
20 o3 89.2% high Apr 2025 Official leaderboard
21 DeepSeek-R1 May 2025 89.2% May 2025 Official leaderboard
22 gpt-oss-20b 89.2% high Aug 2025 Official leaderboard
23 Grok 4.1 Fast 89.2% reasoning Nov 2025 Official leaderboard
24 Gemini 2.5 Pro Mar 2025 88.3% Mar 2025 Official leaderboard
25 GPT-5 mini 87.5% high Aug 2025 Official leaderboard
26 o3-mini 86.7% high Jan 2025 Official leaderboard
27 Falcon H1R 7B 86.7% Jan 2026 Official leaderboard
28 GPT-5 nano 85% high Aug 2025 Official leaderboard
29 Claude Sonnet 4.5 84.2% thinking Sep 2025 Official leaderboard
30 Gemini 2.5 Pro May 2025 83.3% May 2025 Official leaderboard
31 GLM-4.5-Air 83.3% Jul 2025 Official leaderboard
32 K2-Think 83.3% Sep 2025 Official leaderboard
33 o1 81.7% medium Dec 2024 Official leaderboard
34 Grok-3 mini 81.7% high Apr 2025 Official leaderboard
35 Qwen3-235B-A22B 80.8% Apr 2025 Official leaderboard
36 QED-Nano 77.5% Feb 2026 Official leaderboard
37 Gemini 2.5 Flash Sep 2025 70.8% thinking Sep 2025 Official leaderboard
38 DeepSeek-R1 70% Jan 2025 Official leaderboard
39 Qwen3-30B-A3B 70% Apr 2025 Official leaderboard
40 Claude Opus 4 70% thinking May 2025 Official leaderboard
41 QwQ-32B 65.8% Mar 2025 Official leaderboard
42 DeepSeek-R1-Distill-Qwen-32B 60% Jan 2025 Official leaderboard
43 DeepSeek-R1-Distill-Llama-70B 55% Jan 2025 Official leaderboard
44 Gemini 2.0 Flash Thinking Jan 2025 53.3% Jan 2025 Official leaderboard
45 DeepSeek-V3 Mar 2025 50% Mar 2025 Official leaderboard
46 DeepSeek-R1-Distill-Qwen-14B 49.2% Jan 2025 Official leaderboard
47 Claude 3.7 Sonnet 49.2% thinking Feb 2025 Official leaderboard
48 QwQ-32B-Preview 33.3% Nov 2024 Official leaderboard
49 Gemini 2.0 Flash Feb 2025 27.5% Feb 2025 Official leaderboard
50 Gemini 2.0 Pro 27.5% Feb 2025 Official leaderboard
51 DeepSeek-R1-Distill-Qwen-1.5B 20% Jan 2025 Official leaderboard
52 GPT-4o Mar 2025 11.7% Mar 2025 Official leaderboard
53 Claude 3.5 Sonnet 3.33% Jun 2024 Official leaderboard
#ModelScoreSettingDateSource
1 Grok 4 Heavy 100% Jul 2025 Lab model card
2 GPT-5 100% Aug 2025 Lab model card
3 Qwen3-Max 100% Sep 2025 Lab model card
4 Claude Sonnet 4.5 100% Sep 2025 Lab model card
5 Gemini 3 Pro 100% Nov 2025 Lab model card
6 GPT-5.2 100% high Dec 2025 Official leaderboard
7 GPT-5.2 Pro 100% Dec 2025 Lab model card
8 LongCat-Flash-Thinking-2601 100% Jan 2026 Lab model card
9 Claude Opus 4.6 99.8% Feb 2026 Lab model card
10 Gemini 3 Flash 99.7% Dec 2025 Lab model card
11 Gemini 2.5 Deep Think 99.2% Aug 2025 Lab model card
12 Grok 4 98.8% Jul 2025 Lab model card
13 gpt-oss-20b 98.7% Aug 2025 Lab model card
14 Step 3.5 Flash 98.3% Feb 2026 Official leaderboard
15 Seed2.0 Pro 98.3% Feb 2026 Lab model card
16 gpt-oss-120b 97.9% Aug 2025 Lab model card
17 MAI-Thinking-1 97% Jun 2026 Lab model card
18 GLM-5 96.7% Feb 2026 Official leaderboard
19 Claude Haiku 4.5 96.3% Oct 2025 Lab model card
20 Trinity Large Thinking 96.3% Apr 2026 Lab model card
21 Kimi K2.5 96.1% Jan 2026 Lab model card
22 DeepSeek-V3.2-Speciale 96% Dec 2025 Lab model card
23 GLM-4.7 95.7% Dec 2025 Lab model card
24 Claude Sonnet 4.6 95.6% Feb 2026 Lab model card
25 Kimi K2 Thinking 94.5% Nov 2025 Lab model card
26 Seed1.8 94.3% Dec 2025 Lab model card
27 GPT-5.1 94.2% high Nov 2025 Official leaderboard
28 DeepSeek-V3.2 94.2% thinking Dec 2025 Official leaderboard
29 MiMo-V2-Flash 94.1% Dec 2025 Lab model card
30 MiniMax-M2.7 94% May 2026 Lab model card
31 GLM-4.6 93.9% Sep 2025 Lab model card
32 Ring-1T 93.4% Oct 2025 Lab model card
33 GLM-4.5 93.3% Aug 2025 Official leaderboard
34 Grok 3 93.3% Feb 2025 Lab model card
35 Intern-S1-Pro 93.1% Feb 2026 Lab model card
36 EXAONE 4.5 33B 92.9% Apr 2026 Lab model card
37 K-EXAONE 92.8% Dec 2025 Lab model card
38 Claude Opus 4.5 92.8% Nov 2025 Lab model card
39 o4-mini 92.7% Apr 2025 Lab model card
40 Nemotron-Cascade 2 30B-A3B 92.4% Mar 2026 Lab model card
41 Qwen3-235B-A22B-Thinking Jul 2025 92.3% Jul 2025 Lab model card
42 Nova 2.0 Pro Preview 92.3% Dec 2025 Lab model card
43 Nova 2 Omni Preview 92.1% Dec 2025 Lab model card
44 Grok 4 Fast 92% Sep 2025 Lab model card
45 DeepSeek-V3.2-Exp 91.7% thinking Sep 2025 Official leaderboard
46 glm-4.7-flash 91.6% Jan 2026 Lab model card
47 GPT-5 mini 91.1% Aug 2025 Lab model card
48 Nova 2 Lite 91% Dec 2025 Lab model card
49 DeepSeek-V3.1 90.8% thinking Aug 2025 Official leaderboard
50 Grok-3 mini 90.8% Feb 2025 Lab model card
51 LongCat-Flash-Thinking 90.6% Sep 2025 Lab model card
52 Nemotron 3 Super 90.2% Mar 2026 Lab model card
53 Claude Opus 4 90% May 2025 Lab model card
54 Command A+ 90% May 2026 Lab model card
55 Qwen3-VL-235B-A22B 89.7% Sep 2025 Lab model card
56 Cogito v2.1 89.5% Nov 2025 Lab model card
57 o3 89.2% high Apr 2025 Official leaderboard
58 DeepSeek-R1 May 2025 89.2% May 2025 Official leaderboard
59 Grok 4.1 Fast 89.2% reasoning Nov 2025 Official leaderboard
60 Granite 4.2 30B 89.2% Aug 2026 Lab model card
61 Nemotron 3 Nano 30B-A3B 89.1% Dec 2025 Lab model card
62 ERNIE 5.0 89.1% Feb 2026 Lab model card
63 Gemini 2.5 Pro Mar 2025 88.3% Mar 2025 Official leaderboard
64 Sarvam 105B 88.3% Mar 2026 Lab model card
65 Gemini 2.5 Pro Jun 2025 88% Jun 2025 Lab model card
66 INTELLECT-3 88% Nov 2025 Lab model card
67 Qwen3-Next-80B-A3B 87.8% Sep 2025 Lab model card
68 STEP3-VL-10B 87.7% Jan 2026 Lab model card
69 Keye-VL-2.0-30B-A3B 86.7% May 2026 Lab model card
70 o3-mini 86.7% high Jan 2025 Official leaderboard
71 Falcon H1R 7B 86.7% Jan 2026 Official leaderboard
72 MiniMax-M2.5 86.3% Feb 2026 Lab model card
73 Mistral Medium 3.5 86.3% May 2026 Lab model card
74 Seed1.6-Thinking 86% Jun 2025 Lab model card
75 Intern-S1 86% Jul 2025 Lab model card
76 EXAONE 4.0 32B 85.3% Jul 2025 Lab model card
77 GPT-5 nano 85.2% Aug 2025 Lab model card
78 Claude Sonnet 4 85% May 2025 Lab model card
79 Ministral 3 14B 85% Dec 2025 Lab model card
80 seed-oss-36b-instruct 84.7% Aug 2025 Lab model card
81 Solar Open 100B 84.3% Dec 2025 Lab model card
82 Mistral Small 4 83.8% Mar 2026 Lab model card
83 Magistral Medium 1.2 83.5% Sep 2025 Lab model card
84 Gemini 2.5 Pro May 2025 83.3% May 2025 Official leaderboard
85 GLM-4.5-Air 83.3% Jul 2025 Official leaderboard
86 K2-Think 83.3% Sep 2025 Official leaderboard
87 Magistral Medium 1.0 83.3% Jun 2025 Lab model card
88 Qwen3-Coder-Next 83.1% Feb 2026 Lab model card
89 MiniMax-M2.1 83% Dec 2025 Lab model card
90 Step 3 82.9% Jul 2025 Lab model card
91 Llama-3.3-Nemotron-Super-49B-v1.5 82.7% Jul 2025 Lab model card
92 Nemotron 3 Nano Omni 30B-A3B 82.1% Apr 2026 Lab model card
93 o1 81.7% medium Dec 2024 Official leaderboard
94 Qwen3-235B-A22B 81.5% Apr 2025 Lab model card
95 GPT-5.5 Instant 81.2% May 2026 Lab model card
96 Solar Pro 3 80.3% Jan 2026 Lab model card
97 Sarvam 30B 80% Mar 2026 Lab model card
98 Nemotron 3 Nano 4B 78.5% Mar 2026 Lab model card
99 Hermes 4 405B 78.1% Aug 2025 Lab model card
100 Olmo 3.1 32B Think 78.1% Dec 2025 Lab model card
101 ERNIE-4.5-21B-A3B-Thinking 78% Sep 2025 Lab model card
102 Gemini 2.5 Flash Apr 2025 78% Apr 2025 Lab model card
103 Phi-4-Reasoning-plus 78% Apr 2025 Lab model card
104 Claude Opus 4.1 78% Aug 2025 Lab model card
105 MiniMax-M2 78% Oct 2025 Lab model card
106 QED-Nano 77.5% Feb 2026 Official leaderboard
107 Magistral Small 1.2 77.3% Sep 2025 Lab model card
108 MiniMax-M1-80k 76.9% Jun 2025 Lab model card
109 Hunyuan-A13B-Instruct 76.8% Jun 2025 Lab model card
110 InternVL3.5-241B-A28B 75.6% Aug 2025 Lab model card
111 Gemini 2.5 Flash Sep 2025 75.6% Sep 2025 Lab model card
112 Seed-Thinking-v1.5 74% Apr 2025 Lab model card
113 Qwen3-32B 72.9% May 2025 Lab model card
114 Baichuan-M2-32B 72.9% Aug 2025 Lab model card
115 Llama-3.1-Nemotron-Ultra-253B-v1 72.5% Apr 2025 Lab model card
116 Olmo 3 32B Think 72.5% Nov 2025 Lab model card
117 nvidia-nemotron-nano-9b-v2 72.1% Aug 2025 Lab model card
118 Gemini 2.5 Flash Jun 2025 72% Jun 2025 Lab model card
119 Nemotron Nano 12B v2 VL 71.3% Nov 2025 Lab model card
120 Qwen3-30B-A3B 70.9% May 2025 Lab model card
121 Qwen3-14B 70.4% May 2025 Lab model card
122 DeepSeek-R1 70% Jan 2025 Official leaderboard
123 Hermes 4.3 36B 69.3% Dec 2025 Lab model card
124 Solar Pro 2 69.1% Jul 2025 Lab model card
125 Code World Model (CWM) 68.2% Sep 2025 Lab model card
126 Qwen3-8B 67.3% May 2025 Lab model card
127 QwQ-32B 65.8% Mar 2025 Official leaderboard
128 EXAONE Deep 32B 65.8% Mar 2025 Lab model card
129 Qwen3-4B 65.6% May 2025 Lab model card
130 Qwen3-Omni-30B-A3B 65% Sep 2025 Lab model card
131 INTELLECT-2 64.9% May 2025 Lab model card
132 GLM-Z1-32B-0414 63.6% Apr 2025 Lab model card
133 Phi-4-reasoning 63.1% Apr 2025 Lab model card
134 Gemini 2.5 Flash-Lite Jun 2025 63.1% Jun 2025 Lab model card
135 Magistral Small 1.0 62.8% Jun 2025 Lab model card
136 Magistral Small 1.1 62% Jul 2025 Lab model card
137 LongCat-Flash-Chat 61.3% Aug 2025 Lab model card
138 Magistral Medium 1.1 61% Jul 2025 Lab model card
139 DeepSeek-R1-Distill-Qwen-32B 60% Jan 2025 Official leaderboard
140 Llama-3.3-Nemotron-Super-49B-v1 60% May 2025 Lab model card
141 MiMo-7B-RL 55.4% Apr 2025 Lab model card
142 DeepSeek-R1-Distill-Llama-70B 55% Jan 2025 Official leaderboard
143 Gemini 2.0 Flash Thinking Jan 2025 53.3% Jan 2025 Official leaderboard
144 LFM2.5-2.6B 51.9% Aug 2026 Lab model card
145 Reka Flash 3 50% Mar 2025 Lab model card
146 DeepSeek-V3 Mar 2025 50% Mar 2025 Official leaderboard
147 Kimi K2 Jul 2025 49.5% Jul 2025 Lab model card
148 DeepSeek-R1-Distill-Qwen-14B 49.2% Jan 2025 Official leaderboard
149 Claude 3.7 Sonnet 49.2% thinking Feb 2025 Official leaderboard
150 Gemini 2.5 Flash-Lite Sep 2025 48.2% Sep 2025 Lab model card
151 LFM2.5-8B-A1B 42.5% May 2026 Lab model card
152 Qwen3-1.7B 36.8% May 2025 Lab model card
153 ERNIE-4.5-300B-A47B 35.1% Jun 2025 Lab model card
154 Phi-4-mini-flash-reasoning 33.6% Jul 2025 Lab model card
155 QwQ-32B-Preview 33.3% Nov 2024 Official leaderboard
156 LFM2.5-1.2B-Thinking 31.7% Jan 2026 Lab model card
157 Gemini 2.0 Flash Feb 2025 27.5% Feb 2025 Official leaderboard
158 Gemini 2.0 Pro 27.5% Feb 2025 Official leaderboard
159 Trinity Large Preview 24.4% Feb 2026 Lab model card
160 DeepSeek-R1-Distill-Qwen-1.5B 20% Jan 2025 Official leaderboard
161 Falcon-H1-34B 16.7% May 2025 Lab model card
162 Amazon Nova Premier 16% Apr 2025 Lab model card
163 LFM2.5-1.2B-Instruct 14% Jan 2026 Lab model card
164 GPT-4o Mar 2025 11.7% Mar 2025 Official leaderboard
165 Gemma 3n E4B 11.6% Jun 2025 Lab model card
166 Claude 3.5 Sonnet 3.33% Jun 2024 Official leaderboard

MathArena’s leaderboard has 53 models. Switch the scores above to rank them together.

The best score, 100%, costs $0.11 a task

Runs on the cost frontier score higher than any run that's cheaper. DeepSeek-V3.2-Speciale comes within 4.2 points of the best for $0.009 a task, 12 times cheaper. The costs are the leaderboard's own estimates, so compare them only within this benchmark.

xAI OpenAI DeepSeek Technology Innovation Institute StepFun Cost frontier One model at each effort level
0% 20% 40% 60% 80% 100% $0.001 $0.01 $0.10 $1.00 $10 Cost per task, log scale GPT-5.2, high Step 3.5 Flash DeepSeek-V3.2, thinking Falcon H1R 7B Grok-3 mini, low GPT-5 nano, low 0% 20% 40% 60% 80% 100% $0.001 $0.01 $0.10 $1.00 $10 Cost per task, log scale GPT-5.2, high DeepSeek-V3.2, thinking

Source: MathArena’s leaderboard

ModelEffortHarnessScoreCost per taskOutput tokens
GPT-5.2 high 100% $0.11 7,758
GPT-5.2 xhigh 100% $0.18 12,811
Step 3.5 Flash 98.3% $0.01 37,760
Gemini 3 Flash high 97.5% $0.06 18,430
GLM-5 96.7% $0.08 25,259
DeepSeek-V3.2-Speciale 95.8% $0.009 21,957
Kimi K2.5 high 95.8% $0.07 22,716
GPT-5 high 95% $0.14 13,475
Gemini 3 Pro 95% $0.18 14,799
GPT-5.1 high 94.2% $0.18 17,912
DeepSeek-V3.2 thinking 94.2% $0.006 14,908
GLM-4.5 93.3% $0.05 21,965
Grok 4 92.5% $0.19 12,873
Kimi K2 Thinking 92.5% $0.06 24,036
o4-mini high 91.7% $0.06 13,982
DeepSeek-V3.2-Exp thinking 91.7% $0.006 14,024
GLM-4.6 91.7% $0.04 16,491
DeepSeek-V3.1 thinking 90.8% $0.03 14,961
Grok 4 Fast 90.8% $0.005 9,160
gpt-oss-120b high 90% $0.01 21,204
o3 high 89.2% $0.10 11,885
DeepSeek-R1 May 2025 89.2% $0.05 21,923
gpt-oss-20b high 89.2% $0.008 40,223
Grok 4.1 Fast reasoning 89.2% $0.005 10,009
Gemini 2.5 Pro Mar 2025 88.3% $0.13 13,397
GPT-5 mini high 87.5% $0.03 16,431
o3-mini high 86.7% $0.05 11,392
Falcon H1R 7B 86.7% $0.004 17,859
GPT-5 nano high 85% $0.01 27,091
o4-mini medium 84.2% $0.03 6,221
Claude Sonnet 4.5 thinking 84.2% $0.26 17,251
Gemini 2.5 Pro May 2025 83.3% 4,369
GLM-4.5-Air 83.3% $0.03 24,021
K2-Think 83.3%
o1 medium 81.7% $0.71 11,813
Grok-3 mini high 81.7% $0.009 18,460
Qwen3-235B-A22B 80.8% $0.009 14,907
QED-Nano 77.5% 39,431
o3-mini medium 76.7% $0.03 6,182
Gemini 2.5 Flash Sep 2025 thinking 70.8% $0.08 23,871
DeepSeek-R1 70% $0.03 11,211
Qwen3-30B-A3B 70% $0.005 17,970
Claude Opus 4 thinking 70% $1.13 15,044
QwQ-32B 65.8% $0.02 16,133
Grok-3 mini low 65% $0.003 5,707
o4-mini low 61.7% $0.01 2,605
DeepSeek-R1-Distill-Qwen-32B 60% $0.008 12,749
DeepSeek-R1-Distill-Llama-70B 55% $0.006 10,488
Gemini 2.0 Flash Thinking Jan 2025 53.3% 569
DeepSeek-V3 Mar 2025 50% $0.004 3,440
DeepSeek-R1-Distill-Qwen-14B 49.2% $0.004 12,352
Claude 3.7 Sonnet thinking 49.2% $0.37 24,602
o3-mini low 48.3% $0.01 2,361
GPT-5 nano low 47.2% $0.001 2,172
QwQ-32B-Preview 33.3% $0.01 8,085
Gemini 2.0 Flash Feb 2025 27.5% $0.001 2,769
Gemini 2.0 Pro 27.5% $0.01 3,037
DeepSeek-V3 Mar 2025 25% $0.003 2,471
DeepSeek-R1-Distill-Qwen-1.5B 20% $0.003 17,102
GPT-4o Mar 2025 11.7% $0.009 860
Claude 3.5 Sonnet 3.33% $0.009 555

2 models have held number one in labs’ own numbers since its release in February 2025 8 models have held number one across all independent runs since its release in February 2025 3 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 MathArena’s leaderboard and what labs claim in their cards.

OpenAI Google DeepMind xAI Z.ai DeepSeek
MathArena leaderboard Passed 85% in Feb 2025, by o3-mini
GPT-5.2 100%
Lab claims Passed 85% in Feb 2025, by Grok 3
Grok 4 Heavy 100%

Sources: MathArena’s leaderboard and labs’ model cards

O3-mini on the official leaderboard passed 85% in January 2025

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

On MathArena’s 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%) Grok 4 Heavy Grok 3 o3-mini o1 QwQ-32B-Preview Claude 3.5 Sonnet 0% 20% 40% 60% 80% 100% 2024 2025 2026 Near the ceiling (85%) Grok 4 Heavy Grok 3

Sources: MathArena’s leaderboard and labs’ model cards

3 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.

xAI
Reka
NVIDIA
LG AI Research
Google DeepMind
ByteDance Seed
Z.ai
OpenAI
Qwen
Xiaomi Corp
Microsoft
Amazon
Prime Intellect
Technology Innovation Institute
Anthropic
DeepSeek
Mistral AI
MiniMax
Tencent
Baidu
Upstage
Moonshot AI
Shanghai AI Lab
StepFun
Baichuan
Nous Research
Meituan
Meta
Ant Group
Deep Cogito
Ai2
Liquid AI
Arcee AI
Sarvam AI
Cohere
Kuaishou
IBM
DateLaunch postScoreAs the post writes it
Aug 25, 2026 Granite 4.2 30B 89.2 AIME25
Aug 4, 2026 LFM2.5-2.6B 51.9 AIME25
Jun 2, 2026 MAI-Thinking-1 97 AIME 2025
May 28, 2026 LFM2.5-8B-A1B 42.5 AIME25
May 25, 2026 Keye-VL-2.0-30B-A3B 86.7 AIME2025
May 22, 2026 Mistral Medium 3.5 86.3 AIME25 avg@16
May 20, 2026 Command A+ 90 AIME 25 (pass@1 over 10 repeats)
May 5, 2026 GPT-5.5 Instant 81.2 AIME 2025 (Competition Math)
Apr 9, 2026 EXAONE 4.5 33B 92.9 AIME 2025
Apr 1, 2026 Trinity Large Thinking 96.3 AIME25
Mar 19, 2026 Nemotron-Cascade 2 30B-A3B 92.4 AIME 2025 (TIR 98.6)
Mar 16, 2026 Nemotron 3 Nano 4B 78.5 AIME25 (reasoning on)
Mar 16, 2026 Mistral Small 4 83.8 AIME25 [Mistral Small 4 - High]
Mar 11, 2026 Nemotron 3 Super 90.2 AIME25 (no tools)
Mar 6, 2026 Sarvam 30B 80 AIME 25 (80.0 without tools, 96.7 with tools)
Mar 6, 2026 Sarvam 105B 88.3 AIME 25 (88.3 without tools, 96.7 with tools)
Feb 14, 2026 Seed2.0 Pro 98.3 AIME 2025
Feb 12, 2026 MiniMax-M2.5 86.3 AIME25
Feb 6, 2026 ERNIE 5.0 89.1 AIME 2025
Feb 2, 2026 Intern-S1-Pro 93.1 AIME-2025
Feb 2, 2026 Step 3.5 Flash 97.3 AIME 2025
Jan 27, 2026 Trinity Large Preview 24 AIME 2025
Jan 27, 2026 Kimi K2.5 96.1 AIME 2025
Jan 26, 2026 Solar Pro 3 80.3 AIME 2025
Jan 20, 2026 LFM2.5-1.2B-Thinking 31.7 AIME25 (± 1.81)
Jan 19, 2026 glm-4.7-flash 91.6 AIME 25
Jan 14, 2026 LongCat-Flash-Thinking-2601 100 Best of AIME-25 (Avg@16) w/ tools 99.6 / Heavy Thinking 100.0
Jan 13, 2026 STEP3-VL-10B 87.7 AIME 2025 (SeRe)
Jan 5, 2026 Falcon H1R 7B 83.1 AIME-25
Jan 5, 2026 LFM2.5-1.2B-Instruct 14 AIME25
Dec 31, 2025 Solar Open 100B 84.3 AIME 2025
Dec 31, 2025 K-EXAONE 92.8 AIME 2025
Dec 23, 2025 MiniMax-M2.1 83 AIME25
Dec 22, 2025 GLM-4.7 95.7 AIME 2025
Dec 18, 2025 Seed1.8 94.3 AIME-25
Dec 17, 2025 Gemini 3 Flash 99.7 AIME 2025, With code execution; No tools = 95.2%
Dec 16, 2025 MiMo-V2-Flash 94.1 AIME 2025
Dec 15, 2025 Nemotron 3 Nano 30B-A3B 89.1 AIME25 (no tools)
Dec 12, 2025 Olmo 3.1 32B Think 78.1 AIME 2025
Dec 11, 2025 GPT-5.2 Pro 100 AIME 2025 (no tools)
Dec 11, 2025 GPT-5.2 100 AIME 2025 (no tools) Competition math
Dec 3, 2025 Hermes 4.3 36B 69.3 AIME 25
Dec 2, 2025 Ministral 3 14B 85 AIME25 0.850 (reasoning)
Dec 2, 2025 Nova 2 Omni Preview 92.1 AIME 2025 acc
Dec 2, 2025 Nova 2 Lite 91 AIME 2025 acc
Dec 2, 2025 Nova 2.0 Pro Preview 92.3 AIME 2025 acc
Dec 1, 2025 DeepSeek-V3.2 93.1 AIME 2025 (Pass@1)
Nov 26, 2025 INTELLECT-3 88 AIME25
Nov 20, 2025 Olmo 3 32B Think 72.5 AIME 2025
Nov 19, 2025 Cogito v2.1 89.5 AIME 2025
Nov 18, 2025 Gemini 3 Pro 100 AIME 2025, With code execution; No tools = 95.0%
Nov 13, 2025 GPT-5.1 94 AIME 2025 (no tools)
Nov 6, 2025 Kimi K2 Thinking 94.5 AIME25, no tools
Oct 27, 2025 MiniMax-M2 78 AIME25 (AA)
Oct 15, 2025 Claude Haiku 4.5 96.3 High school math competition AIME 2025 (python)
Oct 10, 2025 Ring-1T 93.4 AIME 25 (pass@1)
Sep 30, 2025 GLM-4.6 93.9 AIME 25
Sep 29, 2025 DeepSeek-V3.2-Exp 89.3 AIME 2025
Sep 29, 2025 Claude Sonnet 4.5 100 High school math competition AIME 2025 (python)
Sep 25, 2025 Gemini 2.5 Flash-Lite 48.2 AIME 2025, Thinking (non-thinking 50.1)
Sep 25, 2025 Gemini 2.5 Flash 75.6 AIME 2025 single attempt, model card
Sep 24, 2025 Code World Model (CWM) 68.2 AIME25
Sep 24, 2025 Qwen3-Max 100 AIME 25 (Thinking Heavy w/ Python)
Sep 23, 2025 Qwen3-VL-235B-A22B 89.7 AIME25
Sep 22, 2025 Qwen3-Omni-30B-A3B 65 AIME25
Sep 21, 2025 LongCat-Flash-Thinking 90.6 AIME25(Mean@32)
Sep 19, 2025 Grok 4 Fast 92 AIME 2025 (no tools) pass@1
Sep 17, 2025 Magistral Small 1.2 77.3 AIME25 pass@1
Sep 17, 2025 Magistral Medium 1.2 83.5 AIME25 pass@1
Sep 11, 2025 Qwen3-Next-80B-A3B 87.8 AIME25
Sep 9, 2025 ERNIE-4.5-21B-A3B-Thinking 78 AIME2025 (Avg@32)
Aug 29, 2025 LongCat-Flash-Chat 61.3 AIME25(avg@10)
Aug 26, 2025 Hermes 4 405B 78.1 AIME'25 (reasoning)
Aug 25, 2025 InternVL3.5-241B-A28B 75.6 AIME25
Aug 21, 2025 DeepSeek-V3.1 88.4 AIME 2025 (Pass@1)
Aug 20, 2025 seed-oss-36b-instruct 84.7 AIME25
Aug 18, 2025 nvidia-nemotron-nano-9b-v2 72.1 AIME25
Aug 10, 2025 Baichuan-M2-32B 72.9 AIME25
Aug 7, 2025 GPT-5 nano 85.2 AIME '25 (no tools)
Aug 7, 2025 GPT-5 mini 91.1 AIME '25 (no tools)
Aug 7, 2025 GPT-5 100 AIME 2025 Competition math, GPT-5 pro (python)
Aug 5, 2025 gpt-oss-20b 98.7 AIME 2025 (tools)
Aug 5, 2025 Claude Opus 4.1 78 High school math competition AIME 2025
Aug 5, 2025 gpt-oss-120b 97.9 AIME 2025 (tools)
Aug 1, 2025 Gemini 2.5 Deep Think 99.2 AIME 2025
Jul 31, 2025 Step 3 82.9 AIME25
Jul 25, 2025 Llama-3.3-Nemotron-Super-49B-v1.5 82.7 AIME 2025 (pass@1, avg of 16)
Jul 25, 2025 Qwen3-235B-A22B-Thinking 92.3 AIME25
Jul 24, 2025 Magistral Small 1.1 62 AIME25 pass@1
Jul 24, 2025 Magistral Medium 1.1 61 AIME25 pass@1
Jul 24, 2025 Intern-S1 86 AIME2025
Jul 15, 2025 EXAONE 4.0 32B 85.3 AIME 2025 (reasoning)
Jul 11, 2025 Kimi K2 49.5 AIME 2025, Avg@64
Jul 10, 2025 Solar Pro 2 69.1 AIME'25 pass@1
Jul 9, 2025 Phi-4-mini-flash-reasoning 33.6 AIME25
Jul 9, 2025 Grok 4 98.8 AIME'25, with Python (no tools 91.7)
Jul 9, 2025 Grok 4 Heavy 100 AIME'25 (Competition Math) [Grok 4 Heavy, Python]
Jun 30, 2025 ERNIE-4.5-300B-A47B 35.1 AIME'25
Jun 27, 2025 Hunyuan-A13B-Instruct 76.8 AIME 2025
Jun 26, 2025 Gemma 3n E4B 11.6 AIME 2025 0-shot (IT)
Jun 25, 2025 Seed1.6-Thinking 86 AIME2025
Jun 17, 2025 Gemini 2.5 Flash 72 AIME 2025 single attempt
Jun 17, 2025 Gemini 2.5 Flash-Lite 63.1 AIME 2025, Thinking
Jun 17, 2025 Gemini 2.5 Pro 88 AIME 2025, single attempt (pass@1)
Jun 16, 2025 MiniMax-M1-80k 76.9 AIME 2025
Jun 10, 2025 Magistral Small 1.0 62.8 AIME25 pass@1
Jun 10, 2025 Magistral Medium 1.0 83.3 AIME-25 (maj@64; maj@4 = 72.1, pass@1 = 64.9)
May 28, 2025 DeepSeek-R1 87.5 AIME 2025 (Pass@1)
May 22, 2025 Claude Sonnet 4 85 High school math competition AIME 2025 (parallel test-time compute)
May 22, 2025 Claude Opus 4 90 High school math competition AIME 2025 (parallel test-time compute)
May 20, 2025 Falcon-H1-34B 16.7 AIME-25
May 11, 2025 INTELLECT-2 64.9 AIME25
Apr 30, 2025 Phi-4-reasoning 62.9 AIME 25
Apr 30, 2025 Amazon Nova Premier 16 AIME 2025
Apr 30, 2025 Phi-4-Reasoning-plus 78 AIME 25
Apr 29, 2025 MiMo-7B-RL 55.4 AIME 2025 (Pass@1)
Apr 29, 2025 Qwen3-235B-A22B 81.5 AIME’25
Apr 17, 2025 Gemini 2.5 Flash 78 AIME 2025 single attempt
Apr 16, 2025 o4-mini 92.7 AIME 2025 (no tools)
Apr 16, 2025 o3 88.9 AIME 2025 Competition Math, o3 (no tools)
Apr 14, 2025 GLM-Z1-32B-0414 63.6 AIME25
Apr 10, 2025 Seed-Thinking-v1.5 74 AIME 2025
Apr 7, 2025 Llama-3.1-Nemotron-Ultra-253B-v1 72.5 AIME25 (reasoning on)
Mar 25, 2025 Gemini 2.5 Pro 86.7 AIME 2025, single attempt (pass@1)
Mar 18, 2025 EXAONE Deep 32B 65.8 AIME 2025 pass@1 65.8 (cons@64 80.0)
Mar 18, 2025 Llama-3.3-Nemotron-Super-49B-v1 58.4 AIME25 (reasoning on)
Mar 10, 2025 Reka Flash 3 50 AIME'25 (cons@64)
Feb 19, 2025 Grok-3 mini 90.8 AIME'25 (cons@64; pass@1 = 82.0) [Grok 3 mini Think]
Feb 19, 2025 Grok 3 93.3 AIME'25 (cons@64; pass@1 = 77.3)

81 system cards report it, from 29 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
Jun 9, 2026 Keye-VL-2.0-30B-A3B technical report 86.7 AIME2025 (Figure 5, printed value) Capabilities
Jun 2, 2026 MAI-Thinking-1 technical report 97 AIME 2025, pass@1, 256k max output tokens, avg of 4 runs (Table 11) Capabilities
May 26, 2026 MiniMax-M2 technical report 78 AIME 2025, MiniMax-M2 bar, printed label (Figure 9) Capabilities
May 26, 2026 MiniMax-M2.7 technical report 94 AIME 2025, printed bar label (Figure 9) Capabilities
May 26, 2026 MiniMax-M2.5 technical report 86.3 AIME 2025, printed bar label (Figure 9) Capabilities
Apr 27, 2026 Nemotron 3 Nano Omni 30B-A3B technical report 82.1 AIME25 (no tools), pass@1 average of 8 runs (Table 10) Capabilities
Mar 26, 2026 Intern-S1-Pro technical report 93.1 AIME-2025, thinking (Table 2) Capabilities
Mar 21, 2026 Seed1.8 model card 94.3 AIME-25 (Table 1). Avg@10, 97.3 with increased thinking (Table 7) Capabilities
Mar 19, 2026 Nemotron-Cascade 2 30B-A3B technical report 92.4 AIME 2025, no tools (Table 1). TIR 98.6 Capabilities
Mar 11, 2026 Nemotron 3 Super technical report 90.2 AIME25 (no tools) (Table 5) Capabilities
Feb 28, 2026 Qwen3-Coder-Next technical report 83.1 AIME25 (Table 9) Capabilities
Feb 19, 2026 Trinity Large Preview technical report 24.4 AIME25, Trinity Large Preview (Table 4) Capabilities
Feb 17, 2026 Claude Sonnet 4.6 system card 95.6 AIME 2025, no tools, adaptive thinking, max effort, average of 10 trials (2.10) Capabilities
Feb 14, 2026 Seed2.0 Pro model card 98.3 AIME 2025 (Table 3) Capabilities
Feb 11, 2026 Step 3.5 Flash technical report 97.3 AIME 2025, avg of 64 (Table 5). PaCoRe 99.9, with Python 99.8 (Table 12) Capabilities
Feb 5, 2026 Claude Opus 4.6 system card 99.8 AIME 2025, no tools, adaptive thinking, max effort, average of 5 trials, possible contamination noted (2.12) Capabilities
Feb 4, 2026 ERNIE 5.0 technical report 89.1 AIME 2025 (Table 2) Capabilities
Jan 30, 2026 Kimi K2.5 technical report 96.1 AIME 2025, Avg@64 (Table 4) Capabilities
Jan 23, 2026 LongCat-Flash-Thinking-2601 technical report 100 AIME-25 (Avg@16), w/ tools, heavy mode (Table 2). Default mode 99.6 Capabilities
Jan 14, 2026 STEP3-VL-10B technical report 87.7 AIME2025, SeRe (Tables 2 and 3). PaCoRe 94.43 Capabilities
Jan 13, 2026 Ministral 3 14B technical report 85 AIME 2025, reported as pass@16 (Table 5) Capabilities
Jan 11, 2026 Solar Open 100B technical report 84.3 AIME 2025 (Table 6) Capabilities
Jan 6, 2026 MiMo-V2-Flash technical report 94.1 AIME 2025 (Table 9) Capabilities
Jan 5, 2026 Falcon H1R 7B technical report 83.1 AIME25 pass@1 (Table 4). DeepConf@512 96.7 (Table 7) Capabilities
Jan 5, 2026 K-EXAONE technical report 92.8 AIME 2025 (Table 3, Reasoning mode). Non-reasoning 44.6 (Table 4) Capabilities
Dec 23, 2025 Nemotron 3 Nano 30B-A3B technical report 89.1 AIME25 (no tools) (Table 3). With tools 99.17 Capabilities
Dec 18, 2025 INTELLECT-3 technical report 88 AIME25, Avg@32 (Table 2, Appendix A.1) Capabilities
Dec 15, 2025 Olmo 3 32B Think technical report 72.5 AIME 2025, Final Think 3.0 column (Table 14) Capabilities
Dec 15, 2025 Olmo 3.1 32B Think technical report 78.1 AIME 2025 (Tables 1 and 14) Capabilities
Dec 2, 2025 Nova 2 Omni Preview technical report 92.1 AIME 2025 acc (Table 5) Capabilities
Dec 2, 2025 Nova 2 Lite technical report 91 AIME 2025 acc (Table 1) Capabilities
Dec 2, 2025 Nova 2.0 Pro Preview technical report 92.3 AIME 2025 acc (Table 1) Capabilities
Dec 1, 2025 DeepSeek-V3.2 technical report 93.1 AIME 2025 (Pass@1), DeepSeek-V3.2 Thinking (Table 2) Capabilities
Dec 1, 2025 DeepSeek-V3.2-Speciale technical report 96 AIME 2025 (Pass@1), 23k output tokens, DeepSeek-V3.2-Speciale (Table 3) Capabilities
Nov 26, 2025 Qwen3-VL-235B-A22B technical report 89.7 AIME-25 (Table 6) Capabilities
Nov 24, 2025 Claude Opus 4.5 system card 92.8 AIME 2025, no tools, 64k thinking, avg of 5 trials (2.17; 100% with python tools, possible contamination noted) Capabilities
Nov 6, 2025 Nemotron Nano 12B v2 VL technical report 71.3 AIME-25, reasoning on, SFT stage 4, mean of 16 runs (Table 6) Capabilities
Oct 21, 2025 Ring-1T technical report 93.4 AIME 2025 (Avg@64) (Table 1) Capabilities
Sep 30, 2025 Code World Model (CWM) technical report 68.2 AIME25, pass@1 averaged over 20 samples (Table 12) Capabilities
Sep 29, 2025 DeepSeek-V3.2-Exp technical report 89.3 AIME 2025 (Pass@1), DeepSeek-V3.2-Exp (Table 1) Capabilities
Sep 26, 2025 Gemini 2.5 Flash model card 75.6 AIME 2025, single attempt (pass@1) (Table, p.4) Capabilities
Sep 23, 2025 LongCat-Flash-Thinking technical report 90.6 AIME-25(Mean@32), no tools (Table 2). With tools 90.00 at 6,965 average tokens versus 19,653 (Figure 9) Capabilities
Sep 22, 2025 Qwen3-Omni-30B-A3B technical report 65 AIME25, Instruct (Table 4). Thinking 73.7 (Table 5) Capabilities
Sep 2, 2025 Baichuan-M2-32B technical report 72.9 AIME25, max_tokens 64k (Table 1) Capabilities
Sep 1, 2025 LongCat-Flash-Chat technical report 61.3 AIME25 (avg@10) (Table 3) Capabilities
Aug 25, 2025 InternVL3.5-241B-A28B technical report 75.6 AIME25 (Table 2, Table 14) Capabilities
Aug 25, 2025 Hermes 4 405B technical report 78.1 AIME'25, reasoning mode (Table 3). Non-reasoning 10.6 Capabilities
Aug 21, 2025 Intern-S1 technical report 86 AIME2025, OpenCompass version (Table 2) Capabilities
Aug 20, 2025 nvidia-nemotron-nano-9b-v2 technical report 72 AIME25, NVIDIA-Nemotron-Nano-9B-v2 (Figure 1). The launch model card prints 72.1 Capabilities
Aug 5, 2025 gpt-oss-20b model card 98.7 AIME 2025 (with tools), high reasoning (no tools 91.7) (Table 3, Figure 1) Capabilities
Aug 5, 2025 gpt-oss-120b model card 97.9 AIME 2025 (with tools), high reasoning (no tools 92.5) (Table 3, Figure 1) Capabilities
Jul 30, 2025 Falcon-H1-34B technical report 16.7 AIME-25, average accuracy over 16 repetitions (Table 24) Capabilities
Jul 21, 2025 Kimi K2 technical report 49.5 AIME 2025 (Avg@64) (Table 3) Capabilities
Jul 15, 2025 EXAONE 4.0 32B technical report 85.3 AIME 2025, n=32, 64K budget (Table 3, Reasoning). 32K budget 74.8, 1K 35.6 (Table 7). Non-reasoning 35.9 (Table 4) Capabilities
Jul 9, 2025 Phi-4-mini-flash-reasoning technical report 33.6 AIME25, pass@1 averaged over 64 samples (Table 4) Capabilities
Jul 7, 2025 Gemini 2.5 Pro technical report 88 AIME 2025, single attempt pass@1 (Table 3) Capabilities
Jul 7, 2025 Gemini 2.5 Flash technical report 72 AIME 2025, single attempt pass@1 (Table 3) Capabilities
Jun 29, 2025 ERNIE-4.5-300B-A47B technical report 35.1 AIME'25 (Table 5) Capabilities
Jun 26, 2025 Hunyuan-A13B-Instruct technical report 76.8 AIME2025, slow-thinking (Table 3). Fast-thinking 19.2 (Table 4) Capabilities
Jun 16, 2025 MiniMax-M1-80k technical report 76.9 AIME 2025, avg of 32 samples (Table 2) Capabilities
Jun 12, 2025 Magistral Small 1.0 technical report 62.8 AIME'25 pass@1, SFT + RL (Magistral Small) column (Table 3). maj@64 76.7 Capabilities
Jun 12, 2025 Magistral Medium 1.0 technical report 64.9 AIME'25, pass@1 averaged over 64 runs (Table 2). maj@64 83.3 Capabilities
May 14, 2025 Qwen3-8B technical report 67.3 AIME’25, thinking mode (Table 17). Non-thinking 20.9 (Table 18) Capabilities
May 14, 2025 Qwen3-4B technical report 65.6 AIME’25, thinking mode (Table 17). Non-thinking 19.1 (Table 18) Capabilities
May 14, 2025 Qwen3-32B technical report 72.9 AIME’25, thinking mode (Table 13). Non-thinking 20.2 (Table 14) Capabilities
May 14, 2025 Qwen3-30B-A3B technical report 70.9 AIME’25, thinking mode (Table 15). Non-thinking 21.6 (Table 16) Capabilities
May 14, 2025 Qwen3-235B-A22B technical report 81.5 AIME’25, thinking mode (Table 11). Non-thinking 24.7 (Table 12) Capabilities
May 14, 2025 Qwen3-14B technical report 70.4 AIME’25, thinking mode (Table 15). Non-thinking 23.3 (Table 16) Capabilities
May 14, 2025 Qwen3-1.7B technical report 36.8 AIME’25, thinking mode (Table 19). Non-thinking 9.8 (Table 20) Capabilities
May 12, 2025 MiMo-7B-RL technical report 55.4 AIME 2025 (Pass@1), average of 32 runs (Table 4, Table 5, Table 6) Capabilities
May 12, 2025 INTELLECT-2 technical report 64.9 AIME25, evalchemy defaults, target length 10,000 tokens (Table 1) Capabilities
May 2, 2025 Llama-3.3-Nemotron-Super-49B-v1 technical report 60 AIME25, full 30 questions, reasoning on, reasoning off 16.7 (Table 4). The launch model card prints 58.4 Capabilities
May 2, 2025 Llama-3.1-Nemotron-Ultra-253B-v1 technical report 72.5 AIME25, full 30 questions, reasoning on, reasoning off 16.7 (Table 5) Capabilities
Apr 30, 2025 Phi-4-reasoning technical report 63.1 AIME 25, average pass@1, std 6.3 (Table 1, Figure 8) Capabilities
Apr 30, 2025 Phi-4-Reasoning-plus technical report 78 AIME 25, average pass@1 over 50 runs (Table 1) Capabilities
Apr 30, 2025 Amazon Nova Premier technical report 16 AIME 2025, mean of 5 samples (Table 1) Capabilities
Apr 29, 2025 Gemini 2.5 Flash model card 78 AIME 2025, single attempt (pass@1) (Table, p.4) Capabilities
Apr 28, 2025 Gemini 2.5 Pro model card 86.7 AIME 2025, single attempt pass@1 (Table, p.4, column Gemini 2.5 Pro Experimental 03-25) Capabilities
Apr 10, 2025 Seed-Thinking-v1.5 technical report 74 AIME 2025, mean of 32 samples (Table 2) Capabilities
Mar 16, 2025 EXAONE Deep 32B technical report 65.8 AIME 2025, pass@1 with k=64, cons@64 80.0 (Table 2) Capabilities
Jan 20, 2025 DeepSeek-R1 technical report AIME 2025 average score 11.3/15, USAMO index 256.7 (Table 13) Capabilities

The lab's number comes out ahead in 20 of 29 pairs that differ

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. 1 of them match, which usually means one side copied the other. Every benchmark.

Model Lab says Board says Gap What differs
Claude Opus 4 90% High school math competition AIME 2025 (parallel test-time compute) 70% Leaderboard, thinking +20 pts The lab used parallel test-time compute, while MathArena averages separate individual runs.
Claude Sonnet 4.5 100% High school math competition AIME 2025 (python) 84.2% Leaderboard, thinking +16 pts
gpt-oss-20b 98.7% AIME 2025 (tools) 89.2% Leaderboard, high +9.5 pts
Grok-3 mini 90.8% AIME'25 (cons@64; pass@1 = 82.0) [Grok 3 mini Think] 81.7% Leaderboard, high +9.1 pts
gpt-oss-120b 97.9% AIME 2025 (tools) 90% Leaderboard, high +7.9 pts
Grok 4 98.8% AIME'25, with Python (no tools 91.7) 92.5% Leaderboard +6.3 pts
GPT-5 100% AIME 2025 Competition math, GPT-5 pro (python) 95% Leaderboard, high +5.0 pts
Gemini 3 Pro 100% AIME 2025, With code execution; No tools = 95.0% 95% Leaderboard +5.0 pts
Gemini 2.5 Flash Sep 2025 75.6% AIME 2025 single attempt, model card 70.8% Leaderboard, thinking +4.8 pts
GPT-5 mini 91.1% AIME '25 (no tools) 87.5% Leaderboard, high +3.6 pts
Falcon H1R 7B 83.1% AIME-25 86.7% Leaderboard −3.6 pts
DeepSeek-V3.1 88.4% AIME 2025 (Pass@1) 90.8% Leaderboard, thinking −2.4 pts
DeepSeek-V3.2-Exp 89.3% AIME 2025 91.7% Leaderboard, thinking −2.4 pts
GLM-4.6 93.9% AIME 25 91.7% Leaderboard +2.2 pts
Gemini 3 Flash 99.7% AIME 2025, With code execution; No tools = 95.2% 97.5% Leaderboard, high +2.2 pts
Kimi K2 Thinking 94.5% AIME25, no tools 92.5% Leaderboard +2.0 pts
DeepSeek-R1 May 2025 87.5% AIME 2025 (Pass@1) 89.2% Leaderboard −1.7 pts
Gemini 2.5 Pro Mar 2025 86.7% AIME 2025, single attempt (pass@1) 88.3% Leaderboard −1.6 pts
Grok 4 Fast 92% AIME 2025 (no tools) pass@1 90.8% Leaderboard +1.2 pts
DeepSeek-V3.2 93.1% AIME 2025 (Pass@1) 94.2% Leaderboard, thinking −1.1 pts
o4-mini 92.7% AIME 2025 (no tools) 91.7% Leaderboard, high +1.0 pts
Step 3.5 Flash 97.3% AIME 2025 98.3% Leaderboard −1.0 pts
Qwen3-30B-A3B 70.9% AIME’25, thinking mode (Table 15). Non-thinking 21.6 (Table 16) 70% Leaderboard +0.9 pts
Qwen3-235B-A22B 81.5% AIME’25 80.8% Leaderboard +0.7 pts
o3 88.9% AIME 2025 Competition Math, o3 (no tools) 89.2% Leaderboard, high −0.3 pts
Kimi K2.5 96.1% AIME 2025 95.8% Leaderboard, high +0.3 pts
GPT-5 nano 85.2% AIME '25 (no tools) 85% Leaderboard, high +0.2 pts
GPT-5.1 94% AIME 2025 (no tools) 94.2% Leaderboard, high −0.2 pts
DeepSeek-V3.2-Speciale 96% AIME 2025 (Pass@1), 23k output tokens, DeepSeek-V3.2-Speciale (Table 3) 95.8% Leaderboard +0.2 pts
GPT-5.2 100% AIME 2025 (no tools) Competition math 100% Leaderboard, high Same

Versions

We switch the top row to each new version when it comes out. Open it for one row per version. An old version stays in play while people still test it, so whoever first takes it past 85% gets credit for closing it out.

OpenAI DeepSeek xAI
AIME 2026 Passed 85% in Feb 2026, by GPT-5.2
GPT-5.5 100%
AIME 2024 Passed 85% in Jan 2025, by o3-mini
gpt-oss-120b 96.6%
AIME 2025 Passed 85% in Feb 2025, by Grok 3
Grok 4 Heavy 100%
AIME 2026 Passed 85% in Feb 2026, by GPT-5.2
GPT-5.5 100%

Sources: Official leaderboards and labs’ model cards

Replaces AIME 2024 Replaces MATH-500 Replaced by AIME 2026

More in Math

All 113 in Math
  • IMO-AnswerBench In use 400 olympiad problems with short verifiable answers, robustified against memorization. Scored by answer match. GLM-5.2 91%
  • HMMT 2026 In use Models solve the 33 final-answer problems from the February 2026 Harvard-MIT Mathematics Tournament. Scored by exact answer match. GPT-5.5 98.5%
  • PolyMath In use Models solve 9,000 parallel math problems in 18 languages across four difficulty levels, from school math to frontier math. Final answers are checked. Qwen3.7-Max 86.5%
  • MathArena (Apex, IMO, USAMO) In use Evaluates models on newly released math competitions after release. Final-answer or human/LLM-judged proof scores. Claude Opus 4.8 81.3%
  • MathArena Apex Shortlist In use Models solve 47 hard final-answer problems that MathArena selected from 2025 competitions, where the best models scored about 50% at selection. Scored by accuracy over repeated attempts. GPT-5.5 98.4%
  • FrontierMath Tiers 1-3 v2 In use Models solve 295 unpublished math problems from undergraduate to advanced graduate level and submit answers that code checks automatically. Scored by share correct. Claude Fable 5.2 95.1%