Benchmark Almanac

PolyMath

Qwen3.7-Max 86.5% leads since May 2026
Qwen3-235B-A22B 54.6% leads since April 2025
Qwen3.7-Max 86.5% leads since May 2026
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.

Some model cards report only the English subset, PolyMath-en.

Released
April 2025
Kind
Competition problems with answers
Level
Competition
Scored
Checked automatically
Answers
Short answer
Tasks
All public
Strongest model tested
Qwen3.7-Max #49 on the capability index
Last in a launch post
July 2026
Last model added
August 2026
Scores refreshed
November 12th, 2026

In the authors’ words

In this paper, we introduce PolyMath, a multilingual mathematical reasoning benchmark covering 18 languages and 4 easy-to-hard difficulty levels. Our benchmark ensures difficulty comprehensiveness, language diversity, and high-quality translation, making it a highly discriminative multilingual mathematical benchmark in the era of reasoning LLMs. We conduct a comprehensive evaluation for advanced LLMs and find that even Qwen-3-235B-A22B-Thinking and Gemini-2.5-pro, achieve only 54.6 and 52.2 benchmark scores, with about 40% accuracy under the highest level From a language perspective, our benchmark reveals several key challenges of LLMs in multilingual reasoning: (1) Reasoning performance varies widely across languages for current LLMs; (2) Input-output language consistency is low in reasoning LLMs and may be correlated with performance; (3) The thinking length differs significantly by language for current LLMs. Additionally, we demonstrate that controlling the output language in the instructions has the potential to affect reasoning performance, especially for some low-resource languages, suggesting a promising direction for improving multilingual capabilities in LLMs.

Abstract of PolyMath: Evaluating Mathematical Reasoning in Multilingual Contexts (arXiv:2504.18428, CC0)
24
Models scored in lab cards
17
Launch posts list it, from 3 labs
11
System cards report it, from 3 labs

Qwen3.7-Max leads with 86.5% among models released by July 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.

Qwen3-235B-A22B leads with 54.6% across all independent runs

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

Qwen3.7-Max leads with 86.5%, 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 July 2026, and nobody has reported a newer one. Among the 20 highest-ranked models on the capability index, these have no score yet:

24 models. Scroll sideways for the rest.

  1. 86.5 Qwen3.7-Max
  2. 84.0 Qwen3.7-Plus
  3. 77.4 Qwen 3.6 Plus
  4. 73.3 Qwen3.5 397B-A17B
  5. 71.3 K-EXAONE 2.0
  6. 71.2 Qwen3.5-27B
  7. 68.9 Qwen3.5-122B-A10B
  8. 65.1 Kimi K2
  9. 64.4 Qwen3.5-35B-A3B
  10. 60.1 Qwen3-235B-A22B-Thinking
  11. 57.8 Qwen3-VL-235B-A22B
  12. 57.3 Qwen3.5-9B
  13. 56.3 Qwen3-Next-80B-A3B
  14. 54.7 Qwen3-235B-A22B
  15. 51.1 Qwen3.5-4B
  16. 47.4 Qwen3-32B
  17. 46.1 Qwen3-30B-A3B
  18. 45.8 Qwen3-14B
  19. 42.7 Qwen3-8B
  20. 40.0 Qwen3-4B
  21. 37.9 Qwen3-Omni-30B-A3B
  22. 26.1 Qwen3.5-2B
  23. 25.2 Qwen3-1.7B
  24. 8.20 Qwen3.5-0.8B
  1. 1 Qwen3.7-Max 86.5%
  2. 2 Qwen3.7-Plus 84%
  3. 3 Qwen 3.6 Plus 77.4%
  4. 4 Qwen3.5 397B-A17B 73.3%
  5. 5 K-EXAONE 2.0 71.3%
  6. 6 Qwen3.5-27B 71.2%
  7. 7 Qwen3.5-122B-A10B 68.9%
  8. 8 Kimi K2 Jul 2025 65.1%
  9. 9 Qwen3.5-35B-A3B 64.4%
  10. 10 Qwen3-235B-A22B-Thinking Jul 2025 60.1%
  11. 11 Qwen3-VL-235B-A22B 57.8%
  12. 12 Qwen3.5-9B 57.3%
  13. 13 Qwen3-Next-80B-A3B 56.3%
  14. 14 Qwen3-235B-A22B 54.7%
  15. 15 Qwen3.5-4B 51.1%
  16. 16 Qwen3-32B 47.4%
  17. 17 Qwen3-30B-A3B 46.1%
  18. 18 Qwen3-14B 45.8%
  19. 19 Qwen3-8B 42.7%
  20. 20 Qwen3-4B 40%
  21. 21 Qwen3-Omni-30B-A3B 37.9%
  22. 22 Qwen3.5-2B 26.1%
  23. 23 Qwen3-1.7B 25.2%
  24. 24 Qwen3.5-0.8B 8.20%

Source: Labs’ model cards

38 models. Scroll sideways for the rest.

  1. 54.6 Qwen3-235B-A22B
  2. 52.2 Gemini 2.5 Pro
  3. 47.4 Qwen3-32B
  4. 47.2 Seed-Thinking-v1.5
  5. 47.0 DeepSeek-R1
  6. 46.1 Qwen3-30B-A3B
  7. 45.9 QwQ-32B
  8. 45.8 Qwen3-14B
  9. 45.6 o4-mini
  10. 42.7 Qwen3-8B
  11. 40.0 Qwen3-4B
  12. 38.9 o1
  13. 38.6 o3-mini
  14. 37.1 Gemini 2.0 Flash Thinking
  15. 36.6 o1-mini
  16. 35.2 Grok 3
  17. 35.1 DeepSeek-R1-Distill-Qwen-32B
  18. 33.5 Claude 3.7 Sonnet
  19. 32.3 Grok-3 mini
  20. 30.7 DeepSeek-V3
  21. 29.4 DeepSeek-R1-Distill-Llama-70B
  22. 28.8 GPT-4.1 mini
  23. 28.6 DeepSeek-R1-Distill-Qwen-14B
  24. 28.3 DeepSeek-R1-Distill-Qwen-7B
  25. 26.9 GPT-4.5
  26. 26.4 GPT-4.1
  27. 26.1 Llama 4 Maverick
  28. 25.2 Qwen3-1.7B
  29. 24.3 GPT-4o
  30. 21.0 Qwen2.5-Math-72B-Instruct
  31. 20.9 Llama 4 Scout
  32. 20.4 DeepSeek-V3
  33. 20.3 Gemma 3 27B
  34. 19.4 Qwen2.5-Max
  35. 17.4 Phi-4
  36. 16.9 Qwen2.5-72B
  37. 13.7 GPT-4o
  38. 11.5 Llama 3.3 70B Instruct
  1. 1 Qwen3-235B-A22B thinking 54.6%
  2. 2 Gemini 2.5 Pro Mar 2025 52.2%
  3. 3 Qwen3-32B thinking 47.4%
  4. 4 Seed-Thinking-v1.5 47.2%
  5. 5 DeepSeek-R1 47%
  6. 6 Qwen3-30B-A3B thinking 46.1%
  7. 7 QwQ-32B 45.9%
  8. 8 Qwen3-14B thinking 45.8%
  9. 9 o4-mini medium 45.6%
  10. 10 Qwen3-8B thinking 42.7%
  11. 11 Qwen3-4B thinking 40%
  12. 12 o1 38.9%
  13. 13 o3-mini medium 38.6%
  14. 14 Gemini 2.0 Flash Thinking Jan 2025 37.1%
  15. 15 o1-mini 36.6%
  16. 16 Grok 3 35.2%
  17. 17 DeepSeek-R1-Distill-Qwen-32B 35.1%
  18. 18 Claude 3.7 Sonnet thinking 33.5%
  19. 19 Grok-3 mini 32.3%
  20. 20 DeepSeek-V3 Mar 2025 30.7%
  21. 21 DeepSeek-R1-Distill-Llama-70B 29.4%
  22. 22 GPT-4.1 mini 28.8%
  23. 23 DeepSeek-R1-Distill-Qwen-14B 28.6%
  24. 24 DeepSeek-R1-Distill-Qwen-7B 28.3%
  25. 25 GPT-4.5 26.9%
  26. 26 GPT-4.1 26.4%
  27. 27 Llama 4 Maverick 26.1%
  28. 28 Qwen3-1.7B thinking 25.2%
  29. 29 GPT-4o Mar 2025 24.3%
  30. 30 Qwen2.5-Math-72B-Instruct 21%
  31. 31 Llama 4 Scout 20.9%
  32. 32 DeepSeek-V3 20.4%
  33. 33 Gemma 3 27B 20.3%
  34. 34 Qwen2.5-Max 19.4%
  35. 35 Phi-4 17.4%
  36. 36 Qwen2.5-72B 16.9%
  37. 37 GPT-4o Nov 2024 13.7%
  38. 38 Llama 3.3 70B Instruct 11.5%

Source: Official PolyMath leaderboard

55 models. Scroll sideways for the rest.

  1. 86.5 Qwen3.7-Max
  2. 84.0 Qwen3.7-Plus
  3. 77.4 Qwen 3.6 Plus
  4. 73.3 Qwen3.5 397B-A17B
  5. 71.3 K-EXAONE 2.0
  6. 71.2 Qwen3.5-27B
  7. 68.9 Qwen3.5-122B-A10B
  8. 65.1 Kimi K2
  9. 64.4 Qwen3.5-35B-A3B
  10. 60.1 Qwen3-235B-A22B-Thinking
  11. 57.8 Qwen3-VL-235B-A22B
  12. 57.3 Qwen3.5-9B
  13. 56.3 Qwen3-Next-80B-A3B
  14. 54.7 Qwen3-235B-A22B
  15. 52.2 Gemini 2.5 Pro
  16. 51.1 Qwen3.5-4B
  17. 47.4 Qwen3-32B
  18. 47.2 Seed-Thinking-v1.5
  19. 47.0 DeepSeek-R1
  20. 46.1 Qwen3-30B-A3B
  21. 45.9 QwQ-32B
  22. 45.8 Qwen3-14B
  23. 45.6 o4-mini
  24. 42.7 Qwen3-8B
  25. 40.0 Qwen3-4B
  26. 38.9 o1
  27. 38.6 o3-mini
  28. 37.9 Qwen3-Omni-30B-A3B
  29. 37.1 Gemini 2.0 Flash Thinking
  30. 36.6 o1-mini
  31. 35.2 Grok 3
  32. 35.1 DeepSeek-R1-Distill-Qwen-32B
  33. 33.5 Claude 3.7 Sonnet
  34. 32.3 Grok-3 mini
  35. 30.7 DeepSeek-V3
  36. 29.4 DeepSeek-R1-Distill-Llama-70B
  37. 28.8 GPT-4.1 mini
  38. 28.6 DeepSeek-R1-Distill-Qwen-14B
  39. 28.3 DeepSeek-R1-Distill-Qwen-7B
  40. 26.9 GPT-4.5
  41. 26.4 GPT-4.1
  42. 26.1 Llama 4 Maverick
  43. 26.1 Qwen3.5-2B
  44. 25.2 Qwen3-1.7B
  45. 24.3 GPT-4o
  46. 21.0 Qwen2.5-Math-72B-Instruct
  47. 20.9 Llama 4 Scout
  48. 20.4 DeepSeek-V3
  49. 20.3 Gemma 3 27B
  50. 19.4 Qwen2.5-Max
  51. 17.4 Phi-4
  52. 16.9 Qwen2.5-72B
  53. 13.7 GPT-4o
  54. 11.5 Llama 3.3 70B Instruct
  55. 8.20 Qwen3.5-0.8B
  1. 1 Qwen3.7-Max 86.5%
  2. 2 Qwen3.7-Plus 84%
  3. 3 Qwen 3.6 Plus 77.4%
  4. 4 Qwen3.5 397B-A17B 73.3%
  5. 5 K-EXAONE 2.0 71.3%
  6. 6 Qwen3.5-27B 71.2%
  7. 7 Qwen3.5-122B-A10B 68.9%
  8. 8 Kimi K2 Jul 2025 65.1%
  9. 9 Qwen3.5-35B-A3B 64.4%
  10. 10 Qwen3-235B-A22B-Thinking Jul 2025 60.1%
  11. 11 Qwen3-VL-235B-A22B 57.8%
  12. 12 Qwen3.5-9B 57.3%
  13. 13 Qwen3-Next-80B-A3B 56.3%
  14. 14 Qwen3-235B-A22B 54.7%
  15. 15 Gemini 2.5 Pro Mar 2025 52.2%
  16. 16 Qwen3.5-4B 51.1%
  17. 17 Qwen3-32B thinking 47.4%
  18. 18 Seed-Thinking-v1.5 47.2%
  19. 19 DeepSeek-R1 47%
  20. 20 Qwen3-30B-A3B thinking 46.1%
  21. 21 QwQ-32B 45.9%
  22. 22 Qwen3-14B thinking 45.8%
  23. 23 o4-mini medium 45.6%
  24. 24 Qwen3-8B thinking 42.7%
  25. 25 Qwen3-4B thinking 40%
  26. 26 o1 38.9%
  27. 27 o3-mini medium 38.6%
  28. 28 Qwen3-Omni-30B-A3B 37.9%
  29. 29 Gemini 2.0 Flash Thinking Jan 2025 37.1%
  30. 30 o1-mini 36.6%
  31. 31 Grok 3 35.2%
  32. 32 DeepSeek-R1-Distill-Qwen-32B 35.1%
  33. 33 Claude 3.7 Sonnet thinking 33.5%
  34. 34 Grok-3 mini 32.3%
  35. 35 DeepSeek-V3 Mar 2025 30.7%
  36. 36 DeepSeek-R1-Distill-Llama-70B 29.4%
  37. 37 GPT-4.1 mini 28.8%
  38. 38 DeepSeek-R1-Distill-Qwen-14B 28.6%
  39. 39 DeepSeek-R1-Distill-Qwen-7B 28.3%
  40. 40 GPT-4.5 26.9%
  41. 41 GPT-4.1 26.4%
  42. 42 Llama 4 Maverick 26.1%
  43. 43 Qwen3.5-2B 26.1%
  44. 44 Qwen3-1.7B thinking 25.2%
  45. 45 GPT-4o Mar 2025 24.3%
  46. 46 Qwen2.5-Math-72B-Instruct 21%
  47. 47 Llama 4 Scout 20.9%
  48. 48 DeepSeek-V3 20.4%
  49. 49 Gemma 3 27B 20.3%
  50. 50 Qwen2.5-Max 19.4%
  51. 51 Phi-4 17.4%
  52. 52 Qwen2.5-72B 16.9%
  53. 53 GPT-4o Nov 2024 13.7%
  54. 54 Llama 3.3 70B Instruct 11.5%
  55. 55 Qwen3.5-0.8B 8.20%

Sources: Labs’ model cards and official PolyMath leaderboard

#ModelScoreSettingDateSource
1 Qwen3.7-Max 86.5% May 2026 Lab model card
2 Qwen3.7-Plus 84% Jun 2026 Lab model card
3 Qwen 3.6 Plus 77.4% Apr 2026 Lab model card
4 Qwen3.5 397B-A17B 73.3% Feb 2026 Lab model card
5 K-EXAONE 2.0 71.3% Jul 2026 Lab model card
6 Qwen3.5-27B 71.2% Feb 2026 Lab model card
7 Qwen3.5-122B-A10B 68.9% Feb 2026 Lab model card
8 Kimi K2 Jul 2025 65.1% Jul 2025 Lab model card
9 Qwen3.5-35B-A3B 64.4% Feb 2026 Lab model card
10 Qwen3-235B-A22B-Thinking Jul 2025 60.1% Jul 2025 Lab model card
11 Qwen3-VL-235B-A22B 57.8% Nov 2025 Lab model card
12 Qwen3.5-9B 57.3% Mar 2026 Lab model card
13 Qwen3-Next-80B-A3B 56.3% Sep 2025 Lab model card
14 Qwen3-235B-A22B 54.7% Apr 2025 Lab model card
15 Qwen3.5-4B 51.1% Mar 2026 Lab model card
16 Qwen3-32B 47.4% May 2025 Lab model card
17 Qwen3-30B-A3B 46.1% May 2025 Lab model card
18 Qwen3-14B 45.8% May 2025 Lab model card
19 Qwen3-8B 42.7% May 2025 Lab model card
20 Qwen3-4B 40% May 2025 Lab model card
21 Qwen3-Omni-30B-A3B 37.9% Sep 2025 Lab model card
22 Qwen3.5-2B 26.1% Mar 2026 Lab model card
23 Qwen3-1.7B 25.2% May 2025 Lab model card
24 Qwen3.5-0.8B 8.20% Mar 2026 Lab model card
#ModelScoreSettingDateSource
1 Qwen3-235B-A22B 54.6% thinking Apr 2025 Official leaderboard
2 Gemini 2.5 Pro Mar 2025 52.2% Mar 2025 Official leaderboard
3 Qwen3-32B 47.4% thinking Apr 2025 Official leaderboard
4 Seed-Thinking-v1.5 47.2% Apr 2025 Official leaderboard
5 DeepSeek-R1 47% Jan 2025 Official leaderboard
6 Qwen3-30B-A3B 46.1% thinking Apr 2025 Official leaderboard
7 QwQ-32B 45.9% Mar 2025 Official leaderboard
8 Qwen3-14B 45.8% thinking Apr 2025 Official leaderboard
9 o4-mini 45.6% medium Apr 2025 Official leaderboard
10 Qwen3-8B 42.7% thinking Apr 2025 Official leaderboard
11 Qwen3-4B 40% thinking Apr 2025 Official leaderboard
12 o1 38.9% Dec 2024 Official leaderboard
13 o3-mini 38.6% medium Jan 2025 Official leaderboard
14 Gemini 2.0 Flash Thinking Jan 2025 37.1% Jan 2025 Official leaderboard
15 o1-mini 36.6% Sep 2024 Official leaderboard
16 Grok 3 35.2% Apr 2025 Official leaderboard
17 DeepSeek-R1-Distill-Qwen-32B 35.1% Jan 2025 Official leaderboard
18 Claude 3.7 Sonnet 33.5% thinking Feb 2025 Official leaderboard
19 Grok-3 mini 32.3% Apr 2025 Official leaderboard
20 DeepSeek-V3 Mar 2025 30.7% Mar 2025 Official leaderboard
21 DeepSeek-R1-Distill-Llama-70B 29.4% Jan 2025 Official leaderboard
22 GPT-4.1 mini 28.8% Apr 2025 Official leaderboard
23 DeepSeek-R1-Distill-Qwen-14B 28.6% Jan 2025 Official leaderboard
24 DeepSeek-R1-Distill-Qwen-7B 28.3% Jan 2025 Official leaderboard
25 GPT-4.5 26.9% Feb 2025 Official leaderboard
26 GPT-4.1 26.4% Apr 2025 Official leaderboard
27 Llama 4 Maverick 26.1% Apr 2025 Official leaderboard
28 Qwen3-1.7B 25.2% thinking Apr 2025 Official leaderboard
29 GPT-4o Mar 2025 24.3% Mar 2025 Official leaderboard
30 Qwen2.5-Math-72B-Instruct 21% Sep 2024 Official leaderboard
31 Llama 4 Scout 20.9% Apr 2025 Official leaderboard
32 DeepSeek-V3 20.4% Dec 2024 Official leaderboard
33 Gemma 3 27B 20.3% Mar 2025 Official leaderboard
34 Qwen2.5-Max 19.4% Jan 2025 Official leaderboard
35 Phi-4 17.4% Dec 2024 Official leaderboard
36 Qwen2.5-72B 16.9% Sep 2024 Official leaderboard
37 GPT-4o Nov 2024 13.7% Nov 2024 Official leaderboard
38 Llama 3.3 70B Instruct 11.5% Dec 2024 Official leaderboard
#ModelScoreSettingDateSource
1 Qwen3.7-Max 86.5% May 2026 Lab model card
2 Qwen3.7-Plus 84% Jun 2026 Lab model card
3 Qwen 3.6 Plus 77.4% Apr 2026 Lab model card
4 Qwen3.5 397B-A17B 73.3% Feb 2026 Lab model card
5 K-EXAONE 2.0 71.3% Jul 2026 Lab model card
6 Qwen3.5-27B 71.2% Feb 2026 Lab model card
7 Qwen3.5-122B-A10B 68.9% Feb 2026 Lab model card
8 Kimi K2 Jul 2025 65.1% Jul 2025 Lab model card
9 Qwen3.5-35B-A3B 64.4% Feb 2026 Lab model card
10 Qwen3-235B-A22B-Thinking Jul 2025 60.1% Jul 2025 Lab model card
11 Qwen3-VL-235B-A22B 57.8% Nov 2025 Lab model card
12 Qwen3.5-9B 57.3% Mar 2026 Lab model card
13 Qwen3-Next-80B-A3B 56.3% Sep 2025 Lab model card
14 Qwen3-235B-A22B 54.7% Apr 2025 Lab model card
15 Gemini 2.5 Pro Mar 2025 52.2% Mar 2025 Official leaderboard
16 Qwen3.5-4B 51.1% Mar 2026 Lab model card
17 Qwen3-32B 47.4% thinking Apr 2025 Official leaderboard
18 Seed-Thinking-v1.5 47.2% Apr 2025 Official leaderboard
19 DeepSeek-R1 47% Jan 2025 Official leaderboard
20 Qwen3-30B-A3B 46.1% thinking Apr 2025 Official leaderboard
21 QwQ-32B 45.9% Mar 2025 Official leaderboard
22 Qwen3-14B 45.8% thinking Apr 2025 Official leaderboard
23 o4-mini 45.6% medium Apr 2025 Official leaderboard
24 Qwen3-8B 42.7% thinking Apr 2025 Official leaderboard
25 Qwen3-4B 40% thinking Apr 2025 Official leaderboard
26 o1 38.9% Dec 2024 Official leaderboard
27 o3-mini 38.6% medium Jan 2025 Official leaderboard
28 Qwen3-Omni-30B-A3B 37.9% Sep 2025 Lab model card
29 Gemini 2.0 Flash Thinking Jan 2025 37.1% Jan 2025 Official leaderboard
30 o1-mini 36.6% Sep 2024 Official leaderboard
31 Grok 3 35.2% Apr 2025 Official leaderboard
32 DeepSeek-R1-Distill-Qwen-32B 35.1% Jan 2025 Official leaderboard
33 Claude 3.7 Sonnet 33.5% thinking Feb 2025 Official leaderboard
34 Grok-3 mini 32.3% Apr 2025 Official leaderboard
35 DeepSeek-V3 Mar 2025 30.7% Mar 2025 Official leaderboard
36 DeepSeek-R1-Distill-Llama-70B 29.4% Jan 2025 Official leaderboard
37 GPT-4.1 mini 28.8% Apr 2025 Official leaderboard
38 DeepSeek-R1-Distill-Qwen-14B 28.6% Jan 2025 Official leaderboard
39 DeepSeek-R1-Distill-Qwen-7B 28.3% Jan 2025 Official leaderboard
40 GPT-4.5 26.9% Feb 2025 Official leaderboard
41 GPT-4.1 26.4% Apr 2025 Official leaderboard
42 Llama 4 Maverick 26.1% Apr 2025 Official leaderboard
43 Qwen3.5-2B 26.1% Mar 2026 Lab model card
44 Qwen3-1.7B 25.2% thinking Apr 2025 Official leaderboard
45 GPT-4o Mar 2025 24.3% Mar 2025 Official leaderboard
46 Qwen2.5-Math-72B-Instruct 21% Sep 2024 Official leaderboard
47 Llama 4 Scout 20.9% Apr 2025 Official leaderboard
48 DeepSeek-V3 20.4% Dec 2024 Official leaderboard
49 Gemma 3 27B 20.3% Mar 2025 Official leaderboard
50 Qwen2.5-Max 19.4% Jan 2025 Official leaderboard
51 Phi-4 17.4% Dec 2024 Official leaderboard
52 Qwen2.5-72B 16.9% Sep 2024 Official leaderboard
53 GPT-4o Nov 2024 13.7% Nov 2024 Official leaderboard
54 Llama 3.3 70B Instruct 11.5% Dec 2024 Official leaderboard
55 Qwen3.5-0.8B 8.20% Mar 2026 Lab model card

The benchmark's own leaderboard has 38 models. Switch the scores above to rank them together.

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
Qwen3-235B-A22B
non-thinking thinking
54.6%
Qwen3-32B
non-thinking thinking
47.4%
Qwen3-30B-A3B
non-thinking thinking
46.1%
Qwen3-14B
non-thinking thinking
45.8%
Qwen3-8B
non-thinking thinking
42.7%
Claude 3.7 Sonnet
thinking
33.5%

Source: Official PolyMath leaderboard

ModelEffortHarnessScoreCost per taskOutput tokens
Qwen3-235B-A22B thinking 54.6%
Gemini 2.5 Pro Mar 2025 52.2%
Qwen3-32B thinking 47.4%
Seed-Thinking-v1.5 47.2%
DeepSeek-R1 47%
Qwen3-30B-A3B thinking 46.1%
QwQ-32B 45.9%
Qwen3-14B thinking 45.8%
o4-mini medium 45.6%
Qwen3-8B thinking 42.7%
Qwen3-4B thinking 40%
o1 38.9%
o3-mini medium 38.6%
Gemini 2.0 Flash Thinking Jan 2025 37.1%
o1-mini 36.6%
Grok 3 35.2%
DeepSeek-R1-Distill-Qwen-32B 35.1%
Claude 3.7 Sonnet thinking 33.5%
Grok-3 mini 32.3%
DeepSeek-V3 Mar 2025 30.7%
DeepSeek-R1-Distill-Llama-70B 29.4%
GPT-4.1 mini 28.8%
DeepSeek-R1-Distill-Qwen-14B 28.6%
DeepSeek-R1-Distill-Qwen-7B 28.3%
Qwen3-235B-A22B non-thinking 27%
GPT-4.5 26.9%
GPT-4.1 26.4%
Llama 4 Maverick 26.1%
Qwen3-1.7B thinking 25.2%
GPT-4o Mar 2025 24.3%
Qwen3-30B-A3B non-thinking 22.9%
Qwen3-32B non-thinking 22.5%
Qwen3-14B non-thinking 22%
Qwen2.5-Math-72B-Instruct 21%
Llama 4 Scout 20.9%
DeepSeek-V3 20.4%
Gemma 3 27B 20.3%
Claude 3.7 Sonnet 19.7%
Qwen2.5-Max 19.4%
Qwen3-8B non-thinking 18.8%
Phi-4 17.4%
Qwen2.5-72B 16.9%
GPT-4o Nov 2024 13.7%
Llama 3.3 70B Instruct 11.5%

5 models have held number one in labs’ own numbers since its release in April 2025 2 models have held number one across all independent runs since its release in April 2025 6 models have held number one with claims included since its release in April 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.

Google DeepMind Qwen Moonshot AI
Own leaderboard Not tested since Apr 2025
Qwen3-235B-A22B 54.6%
Lab claims Passed 85% in May 2026, by Qwen3.7-Max
Qwen3.7-Max 86.5%

Sources: Official PolyMath leaderboard and labs’ model cards

Qwen’s own number for Qwen3.7-Max passed 85% in May 2026

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

On the benchmark’s own leaderboard Reported by the lab in its model card Best so far
0% 20% 40% 60% 80% 100% 2025 2026 Near the ceiling (85%) Qwen3.7-Max Qwen 3.6 Plus Kimi K2 Qwen3-235B-A22B DeepSeek-R1 o1 0% 20% 40% 60% 80% 100% 2025 2026 Near the ceiling (85%) Qwen3.7-Max Qwen 3.6 Plus

Sources: Official PolyMath leaderboard and labs’ model cards

2 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
Moonshot AI
LG AI Research
DateLaunch postScoreAs the post writes it
Jul 31, 2026 K-EXAONE 2.0 71.3 PolyMath (nine non-English supported languages)
Jun 1, 2026 Qwen3.7-Plus 84 PolyMATH
May 20, 2026 Qwen3.7-Max 86.5 PolyMATH
Apr 2, 2026 Qwen 3.6 Plus 77.4 PolyMATH
Mar 2, 2026 Qwen3.5-0.8B 8.2 PolyMATH
Mar 2, 2026 Qwen3.5-2B 26.1 PolyMATH
Mar 2, 2026 Qwen3.5-4B 51.1 PolyMATH
Mar 2, 2026 Qwen3.5-9B 57.3 PolyMATH
Feb 24, 2026 Qwen3.5-35B-A3B 64.4 PolyMATH
Feb 24, 2026 Qwen3.5-27B 71.2 PolyMATH
Feb 24, 2026 Qwen3.5-122B-A10B 68.9 PolyMATH
Feb 16, 2026 Qwen3.5 397B-A17B 73.3 PolyMATH
Sep 22, 2025 Qwen3-Omni-30B-A3B 37.9 PolyMATH
Sep 11, 2025 Qwen3-Next-80B-A3B 56.3 PolyMATH
Jul 25, 2025 Qwen3-235B-A22B-Thinking 60.1 PolyMATH
Jul 11, 2025 Kimi K2 65.1 PolyMath-en, Avg@4
Apr 29, 2025 Qwen3-235B-A22B 54.7 PolyMath

11 system cards report it, from 3 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
Aug 5, 2026 K-EXAONE 2.0 technical report 71.3 PolyMath, nine non-English languages (per-language values in Table 7) (Table 6, Reasoning mode) Capabilities
Nov 26, 2025 Qwen3-VL-235B-A22B technical report 57.8 PolyMATH (Table 6) Capabilities
Sep 22, 2025 Qwen3-Omni-30B-A3B technical report 37.9 PolyMATH, Instruct (Table 4). Thinking 47.1 (Table 5) Capabilities
Jul 21, 2025 Kimi K2 technical report 65.1 PolyMath-en (Avg@4) (Table 3) Capabilities
May 14, 2025 Qwen3-8B technical report 42.7 PolyMath, thinking mode (Table 17). Non-thinking 18.8 (Table 18) Capabilities
May 14, 2025 Qwen3-4B technical report 40 PolyMath, thinking mode (Table 17). Non-thinking 16.6 (Table 18) Capabilities
May 14, 2025 Qwen3-32B technical report 47.4 PolyMath, thinking mode (Table 13). Non-thinking 22.5 (Table 14) Capabilities
May 14, 2025 Qwen3-30B-A3B technical report 46.1 PolyMath, thinking mode (Table 15). Non-thinking 23.3 (Table 16) Capabilities
May 14, 2025 Qwen3-235B-A22B technical report 54.7 PolyMath, thinking mode (Table 11). Non-thinking 27.0 (Table 12) Capabilities
May 14, 2025 Qwen3-14B technical report 45.8 PolyMath, thinking mode (Table 15). Non-thinking 22.0 (Table 16) Capabilities
May 14, 2025 Qwen3-1.7B technical report 25.2 PolyMath, thinking mode (Table 19). Non-thinking 10.3 (Table 20) Capabilities

7 lab claims 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. 7 of them match, which usually means one side copied the other. Every benchmark.

Model Lab says Board says Gap
Qwen3-235B-A22B 54.7% PolyMath 54.6% Leaderboard, thinking Same
Qwen3-1.7B 25.2% PolyMath, thinking mode (Table 19). Non-thinking 10.3 (Table 20) 25.2% Leaderboard, thinking Same
Qwen3-14B 45.8% PolyMath, thinking mode (Table 15). Non-thinking 22.0 (Table 16) 45.8% Leaderboard, thinking Same
Qwen3-30B-A3B 46.1% PolyMath, thinking mode (Table 15). Non-thinking 23.3 (Table 16) 46.1% Leaderboard, thinking Same
Qwen3-32B 47.4% PolyMath, thinking mode (Table 13). Non-thinking 22.5 (Table 14) 47.4% Leaderboard, thinking Same
Qwen3-4B 40% PolyMath, thinking mode (Table 17). Non-thinking 16.6 (Table 18) 40% Leaderboard, thinking Same
Qwen3-8B 42.7% PolyMath, thinking mode (Table 17). Non-thinking 18.8 (Table 18) 42.7% Leaderboard, thinking Same

More in Math

All 113 in Math
  • AIME 2026 In use Models solve the 30 integer-answer problems from the 2026 AIME I and II math competitions. Scored by exact answer match. GPT-5.5 100%
  • 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%
  • 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%