Benchmark Almanac

HMMT 2025

GPT-5 100% leads since Aug 2025 Fading

Models answer the 30 problems from the February 2025 Harvard-MIT Mathematics Tournament. Final answers are checked against the official answers.

Model cards use the MathArena version of the February contest, often averaged over many samples.

Made by Harvard-MIT Mathematics Tournament (HMMT) Released Feb 2025 Competition problems with answers All tasks public Last in a launch post Aug 2026 , dropped after it saturated Last model added Aug 2026 Scores refreshed 2 Nov 2026
Level
Competition The strongest students, in olympiad-style contests
Scored
Checked automatically A program checks the result against an answer key, unit tests or the expected final state.
Answers
Short answer A number, word or short string matched to a key.
eth-sri/matharena 294 stars Dataset Leaderboard
49
Models scored in lab cards
44
Launch posts list it, from 19 labs
25
System cards report it, from 16 labs

3 models share the top score of 100% among models released by Aug 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.

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

49 models. Scroll sideways for the rest.

  1. 100.0 GPT-5
  2. 100.0 Qwen3-Max
  3. 100.0 GPT-5.2 Pro
  4. 99.2 DeepSeek-V3.2-Speciale
  5. 98.4 Step 3.5 Flash
  6. 97.9 GLM-5
  7. 97.5 LongCat-Flash-Thinking-2601
  8. 97.3 Seed2.0 Pro
  9. 97.1 GLM-4.7
  10. 96.7 Grok 4 Heavy
  11. 96.7 Qwen 3.6 Plus
  12. 95.4 Kimi K2.5
  13. 94.8 Qwen3.5 397B-A17B
  14. 94.6 Nemotron-Cascade 2 30B-A3B
  15. 93.9 Grok 4
  16. 93.8 Qwen3.6 27B
  17. 93.7 Nemotron 3 Super
  18. 93.3 Grok 4 Fast
  19. 92.5 DeepSeek-V3.2
  20. 91.4 Qwen3.5-122B-A10B
  21. 90.7 Qwen 3.6 35B-A3B
  22. 89.7 Seed1.8
  23. 89.4 Kimi K2 Thinking
  24. 89.2 Granite 4.2 30B
  25. 87.8 GPT-5 mini
  26. 86.7 Ring-1T
  27. 85.8 Tencent HY 2.0 Think
  28. 85.8 Sarvam 105B
  29. 84.4 MiMo-V2-Flash
  30. 84.2 DeepSeek-V3.1
  31. 83.9 Qwen3-235B-A22B-Thinking
  32. 83.7 LongCat-Flash-Thinking
  33. 83.6 DeepSeek-V3.2-Exp
  34. 79.6 ERNIE 5.0
  35. 79.4 DeepSeek-R1
  36. 78.2 STEP3-VL-10B
  37. 77.4 Qwen3-VL-235B-A22B
  38. 75.6 GPT-5 nano
  39. 73.9 Qwen3-Next-80B-A3B
  40. 73.3 Solar Open 100B
  41. 73.3 Sarvam 30B
  42. 72.9 EXAONE 4.0 32B
  43. 70.2 qwen3-coder-next
  44. 70.0 Step 3
  45. 67.5 Ministral 3 14B
  46. 64.9 Falcon H1R 7B
  47. 53.6 Phi-4-Reasoning-plus
  48. 43.8 Phi-4-reasoning
  49. 38.8 Kimi K2
  1. 1 GPT-5 100%
  2. 2 Qwen3-Max 100%
  3. 3 GPT-5.2 Pro 100%
  4. 4 DeepSeek-V3.2-Speciale 99.2%
  5. 5 Step 3.5 Flash 98.4%
  6. 6 GLM-5 97.9%
  7. 7 LongCat-Flash-Thinking-2601 97.5%
  8. 8 Seed2.0 Pro 97.3%
  9. 9 GLM-4.7 97.1%
  10. 10 Grok 4 Heavy 96.7%
  11. 11 Qwen 3.6 Plus 96.7%
  12. 12 Kimi K2.5 95.4%
  13. 13 Qwen3.5 397B-A17B 94.8%
  14. 14 Nemotron-Cascade 2 30B-A3B 94.6%
  15. 15 Grok 4 93.9%
  16. 16 Qwen3.6 27B 93.8%
  17. 17 Nemotron 3 Super 93.7%
  18. 18 Grok 4 Fast 93.3%
  19. 19 DeepSeek-V3.2 92.5%
  20. 20 Qwen3.5-122B-A10B 91.4%
  21. 21 Qwen 3.6 35B-A3B 90.7%
  22. 22 Seed1.8 89.7%
  23. 23 Kimi K2 Thinking 89.4%
  24. 24 Granite 4.2 30B 89.2%
  25. 25 GPT-5 mini 87.8%
  26. 26 Ring-1T 86.7%
  27. 27 Tencent HY 2.0 Think 85.8%
  28. 28 Sarvam 105B 85.8%
  29. 29 MiMo-V2-Flash 84.4%
  30. 30 DeepSeek-V3.1 84.2%
  31. 31 Qwen3-235B-A22B-Thinking Jul 2025 83.9%
  32. 32 LongCat-Flash-Thinking 83.7%
  33. 33 DeepSeek-V3.2-Exp 83.6%
  34. 34 ERNIE 5.0 79.6%
  35. 35 DeepSeek-R1 79.4%
  36. 36 STEP3-VL-10B 78.2%
  37. 37 Qwen3-VL-235B-A22B 77.4%
  38. 38 GPT-5 nano 75.6%
  39. 39 Qwen3-Next-80B-A3B 73.9%
  40. 40 Solar Open 100B 73.3%
  41. 41 Sarvam 30B 73.3%
  42. 42 EXAONE 4.0 32B 72.9%
  43. 43 qwen3-coder-next 70.2%
  44. 44 Step 3 70%
  45. 45 Ministral 3 14B 67.5%
  46. 46 Falcon H1R 7B 64.9%
  47. 47 Phi-4-Reasoning-plus 53.6%
  48. 48 Phi-4-reasoning 43.8%
  49. 49 Kimi K2 Jul 2025 38.8%
#ModelScoreSettingDateSource
1 GPT-5 100% Aug 2025 Model card
2 Qwen3-Max 100% Sep 2025 Model card
3 GPT-5.2 Pro 100% Dec 2025 Model card
4 DeepSeek-V3.2-Speciale 99.2% Dec 2025 Model card
5 Step 3.5 Flash 98.4% Feb 2026 Model card
6 GLM-5 97.9% Feb 2026 Model card
7 LongCat-Flash-Thinking-2601 97.5% Jan 2026 Model card
8 Seed2.0 Pro 97.3% Feb 2026 Model card
9 GLM-4.7 97.1% Dec 2025 Model card
10 Grok 4 Heavy 96.7% Jul 2025 Model card
11 Qwen 3.6 Plus 96.7% Apr 2026 Model card
12 Kimi K2.5 95.4% Jan 2026 Model card
13 Qwen3.5 397B-A17B 94.8% Feb 2026 Model card
14 Nemotron-Cascade 2 30B-A3B 94.6% Mar 2026 Model card
15 Grok 4 93.9% Jul 2025 Model card
16 Qwen3.6 27B 93.8% Apr 2026 Model card
17 Nemotron 3 Super 93.7% Mar 2026 Model card
18 Grok 4 Fast 93.3% Sep 2025 Model card
19 DeepSeek-V3.2 92.5% Dec 2025 Model card
20 Qwen3.5-122B-A10B 91.4% Feb 2026 Model card
21 Qwen 3.6 35B-A3B 90.7% Apr 2026 Model card
22 Seed1.8 89.7% Dec 2025 Model card
23 Kimi K2 Thinking 89.4% Nov 2025 Model card
24 Granite 4.2 30B 89.2% Aug 2026 Model card
25 GPT-5 mini 87.8% Aug 2025 Model card
26 Ring-1T 86.7% Oct 2025 Model card
27 Tencent HY 2.0 Think 85.8% Dec 2025 Model card
28 Sarvam 105B 85.8% Mar 2026 Model card
29 MiMo-V2-Flash 84.4% Dec 2025 Model card
30 DeepSeek-V3.1 84.2% Aug 2025 Model card
31 Qwen3-235B-A22B-Thinking Jul 2025 83.9% Jul 2025 Model card
32 LongCat-Flash-Thinking 83.7% Sep 2025 Model card
33 DeepSeek-V3.2-Exp 83.6% Sep 2025 Model card
34 ERNIE 5.0 79.6% Feb 2026 Model card
35 DeepSeek-R1 79.4% May 2025 Model card
36 STEP3-VL-10B 78.2% Jan 2026 Model card
37 Qwen3-VL-235B-A22B 77.4% Sep 2025 Model card
38 GPT-5 nano 75.6% Aug 2025 Model card
39 Qwen3-Next-80B-A3B 73.9% Sep 2025 Model card
40 Solar Open 100B 73.3% Dec 2025 Model card
41 Sarvam 30B 73.3% Mar 2026 Model card
42 EXAONE 4.0 32B 72.9% Jul 2025 Model card
43 qwen3-coder-next 70.2% Feb 2026 Model card
44 Step 3 70% Jul 2025 Model card
45 Ministral 3 14B 67.5% Jan 2026 Model card
46 Falcon H1R 7B 64.9% Jan 2026 Model card
47 Phi-4-Reasoning-plus 53.6% Apr 2025 Model card
48 Phi-4-reasoning 43.8% Apr 2025 Model card
49 Kimi K2 Jul 2025 38.8% Jul 2025 Model card

The benchmark's own leaderboard has 53 models. We plot every source in the charts below.

The best score, 100%, costs $0.24 a task

Runs on the cost frontier score higher than any run that's cheaper. DeepSeek-V3.2-Speciale comes within 2.5 points of the best for $0.01 a task, 22 times cheaper. The costs are the leaderboard's own estimates, so compare them only within this benchmark.

DeepSeek Google DeepMind xAI OpenAI 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, xhigh Step 3.5 Flash DeepSeek-V3.2, thinking Grok-3 mini, low DeepSeek-R1-Distill-Qwen-14B Gemini 2.0 Flash 0% 20% 40% 60% 80% 100% $0.001 $0.01 $0.10 $1.00 $10 Cost per task, log scale GPT-5.2, xhigh DeepSeek-V3.2, thinking
ModelEffortHarnessScoreCost per taskOutput tokens
GPT-5.2 xhigh 100% $0.24 17,003
GPT-5.2 high 98.3% $0.16 11,164
Step 3.5 Flash 98.3% $0.01 47,820
Gemini 3 Pro 97.5% $0.19 15,918
DeepSeek-V3.2-Speciale 97.5% $0.01 25,870
Gemini 3 Flash 97.5% $0.06 20,536
GLM-5 97.5% $0.09 28,926
Grok 4 95% $0.22 14,669
GLM-4.6 93.3% $0.05 21,480
Kimi K2 Thinking 93.3% $0.07 28,389
GPT-5.1 high 93.3% $0.22 22,001
Kimi K2.5 thinking 93.3% $0.08 26,862
DeepSeek-V3.2 thinking 92.5% $0.007 17,382
Grok 4 Fast 91.7% $0.006 12,770
gpt-oss-120b high 90% $0.02 27,304
DeepSeek-V3.2-Exp thinking 90% $0.008 18,393
Grok 4.1 Fast 90% $0.007 13,404
GPT-5 mini high 89.2% $0.03 16,887
GPT-5 high 88.3% $0.17 16,380
DeepSeek-V3.1 thinking 85.8% $0.04 19,230
Falcon H1R 7B 84.2% $0.006 30,233
o4-mini high 83.3% $0.08 17,637
Gemini 2.5 Pro Mar 2025 82.5% $0.13 12,875
Gemini 2.5 Pro May 2025 80.8% 3,623
o3 high 77.5% $0.12 13,786
GLM-4.5 77.5% $0.06 25,438
DeepSeek-R1 May 2025 76.7% $0.06 25,366
gpt-oss-20b high 76.7% $0.01 54,740
QED-Nano 76.7% 43,216
Grok-3 mini high 74.2% $0.01 21,000
GPT-5 nano high 74.2% $0.01 36,743
GLM-4.5-Air 69.2% $0.03 27,720
o3-mini high 67.5% $0.08 17,660
o4-mini medium 67.5% $0.03 7,298
Claude Sonnet 4.5 thinking 67.5% $0.32 21,410
K2-Think 65%
Gemini 2.5 Flash Sep 2025 thinking 64.2% $0.10 27,168
Qwen3-235B-A22B 62.5% $0.009 15,098
Claude Opus 4 thinking 60% $1.23 16,379
o3-mini medium 53.3% $0.03 7,601
Grok-3 mini low 50.8% $0.003 6,636
Qwen3-30B-A3B 50.8% $0.006 18,539
o1 medium 48.3% $0.89 14,835
QwQ-32B 47.5% $0.02 16,152
o4-mini low 47.5% $0.01 2,663
DeepSeek-R1 41.7% $0.03 12,817
Gemini 2.0 Flash Thinking Jan 2025 35.8% 427
DeepSeek-R1-Distill-Llama-70B 33.3% $0.007 11,898
DeepSeek-R1-Distill-Qwen-32B 33.3% $0.005 15,555
DeepSeek-R1-Distill-Qwen-14B 31.7% $0.002 15,559
Claude 3.7 Sonnet thinking 31.7% $0.39 25,902
DeepSeek-V3 Mar 2025 29.2% $0.005 4,216
o3-mini low 28.3% $0.01 2,661
QwQ-32B-Preview 18.3% $0.01 9,531
Gemini 2.0 Flash Feb 2025 13.3% $0.001 3,312
DeepSeek-V3 Mar 2025 13.3% $0.003 2,453
DeepSeek-R1-Distill-Qwen-1.5B 11.7% $0.004 23,111
Gemini 2.0 Pro 7.50% $0.01 2,252
GPT-4o Mar 2025 5.83% $0.008 769
Claude 3.5 Sonnet 1.67% $0.008 528

6 models have held number one since its release in Feb 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.

OpenAI Google DeepMind xAI Microsoft DeepSeek
Own leaderboard Passed 85% in Jul 2025, by Grok 4
GPT-5.2 100%
Lab claims Passed 85% in Jul 2025, by Grok 4 Heavy
GPT-5 100%

The best score passed 85% in Jul 2025

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% 2024 2025 2026 Near the ceiling (85%) GPT-5 o4-mini o3-mini o1 QwQ-32B-Preview Claude 3.5 Sonnet 0% 20% 40% 60% 80% 100% 2024 2025 2026 Near the ceiling (85%) GPT-5 o4-mini

Labs headline it less often now, last in Aug 2026

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.

DeepSeek
xAI
Moonshot AI
LG AI Research
Qwen
StepFun
OpenAI
Meituan
Ant Group
Tencent
Xiaomi Corp
ByteDance Seed
Z.ai
Upstage
Technology Innovation Institute
Baidu
Sarvam AI
NVIDIA
IBM
DateLaunch postScoreAs the post writes it
25 Aug 2026 Granite 4.2 30B 89.2 HMMT Feb25
22 Apr 2026 Qwen3.6 27B 93.8 HMMT Feb 25
15 Apr 2026 Qwen 3.6 35B-A3B 90.7 HMMT Feb 25
2 Apr 2026 Qwen 3.6 Plus 96.7 HMMT Feb 25
19 Mar 2026 Nemotron-Cascade 2 30B-A3B 94.6 HMMT Feb25
11 Mar 2026 Nemotron 3 Super 93.7 HMMT Feb25 (no tools)
6 Mar 2026 Sarvam 30B 73.3 HMMT Feb 2025
6 Mar 2026 Sarvam 105B 85.8 HMMT (Feb 25)
24 Feb 2026 Qwen3.5-122B-A10B 91.4 HMMT Feb 25
16 Feb 2026 Qwen3.5 397B-A17B 94.8 HMMT Feb 25
14 Feb 2026 Seed2.0 Pro 97.3 HMMT Feb 2025
6 Feb 2026 ERNIE 5.0 79.6 HMMT 2025
2 Feb 2026 Step 3.5 Flash 98.4 HMMT 2025 (Feb.)
27 Jan 2026 Kimi K2.5 95.4 HMMT 2025 (Feb)
23 Jan 2026 Qwen3-Max 98 HMMT Feb 25
14 Jan 2026 LongCat-Flash-Thinking-2601 97.5 Best of HMMT-25 (Avg@16) w/ tools 93.4 / Heavy Thinking 97.5
13 Jan 2026 STEP3-VL-10B 78.2 HMMT 2025 (SeRe)
5 Jan 2026 Falcon H1R 7B 64.9 HMMT-25
31 Dec 2025 Solar Open 100B 73.3 HMMT 2025 (Feb)
22 Dec 2025 GLM-4.7 97.1 HMMT Feb. 2025
18 Dec 2025 Seed1.8 89.7 HMMT25(Feb)
16 Dec 2025 MiMo-V2-Flash 84.4 HMMT Feb. 2025
11 Dec 2025 GPT-5.2 Pro 100 HMMT, Feb 2025 (no tools)
5 Dec 2025 Tencent HY 2.0 Think 85.8 HMMT2025
1 Dec 2025 DeepSeek-V3.2 92.5 HMMT Feb 2025 (Pass@1)
6 Nov 2025 Kimi K2 Thinking 89.4 HMMT25, no tools
10 Oct 2025 Ring-1T 86.7 HMMT 25 (pass@1)
29 Sep 2025 DeepSeek-V3.2-Exp 83.6 HMMT 2025
24 Sep 2025 Qwen3-Max 100 HMMT (Thinking Heavy w/ Python)
23 Sep 2025 Qwen3-VL-235B-A22B 77.4 HMMT25
21 Sep 2025 LongCat-Flash-Thinking 83.7 HMMT25(Mean@32)
19 Sep 2025 Grok 4 Fast 93.3 HMMT 2025 (no tools) pass@1
11 Sep 2025 Qwen3-Next-80B-A3B 73.9 HMMT25
21 Aug 2025 DeepSeek-V3.1 84.2 HMMT 2025 (Pass@1)
7 Aug 2025 GPT-5 nano 75.6 HMMT 2025 (no tools)
7 Aug 2025 GPT-5 mini 87.8 HMMT 2025 (no tools)
7 Aug 2025 GPT-5 100 HMMT Harvard-MIT mathematics tournament, GPT-5 pro (python)
31 Jul 2025 Step 3 70 HMMT25
25 Jul 2025 Qwen3-235B-A22B-Thinking 83.9 HMMT25
15 Jul 2025 EXAONE 4.0 32B 72.9 HMMT Feb 2025 (reasoning)
11 Jul 2025 Kimi K2 38.8 HMMT 2025, Avg@32
9 Jul 2025 Grok 4 93.9 HMMT 2025, with Python (no tools 90.0)
9 Jul 2025 Grok 4 Heavy 96.7 HMMT 2025 (Competitive Math) [Grok 4 Heavy, Python]
28 May 2025 DeepSeek-R1 79.4 HMMT 2025 (Pass@1)

25 system cards report it, from 16 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
21 Mar 2026 Seed1.8 model card 89.7 HMMT25(Feb) (Table 1). Avg@10, 96.7 with increased thinking (Table 7) Capabilities
19 Mar 2026 Nemotron-Cascade 2 30B-A3B technical report 94.6 HMMT Feb25 (Table 1) Capabilities
11 Mar 2026 Nemotron 3 Super technical report 93.7 HMMT Feb25 (no tools), 94.73 with tools (Table 5) Capabilities
28 Feb 2026 qwen3-coder-next technical report 70.2 HMMT25 Feb (Table 9) Capabilities
17 Feb 2026 GLM-5 technical report 97.9 HMMT Feb. 2025 (Table 7) Capabilities
14 Feb 2026 Seed2.0 Pro model card 97.3 HMMT Feb 2025 (Table 3) Capabilities
11 Feb 2026 Step 3.5 Flash technical report 98.4 HMMT 2025 Feb., avg of 64 (Table 5). PaCoRe 100.0, with Python 98.7 (Table 12) Capabilities
4 Feb 2026 ERNIE 5.0 technical report 79.6 HMMT 2025 (Table 2) Capabilities
30 Jan 2026 Kimi K2.5 technical report 95.4 HMMT 2025 (Feb), Avg@64 (Table 4) Capabilities
23 Jan 2026 LongCat-Flash-Thinking-2601 technical report 97.5 HMMT-25 (February) (Avg@16), w/ tools, heavy mode (Table 2). Default mode 93.4 Capabilities
14 Jan 2026 STEP3-VL-10B technical report 78.2 HMMT25, SeRe (Tables 2 and 3). PaCoRe 92.14 Capabilities
13 Jan 2026 Ministral 3 14B technical report 67.5 HMMT 2025, reported as pass@16 (Table 5) Capabilities
11 Jan 2026 Solar Open 100B technical report 73.3 HMMT 2025 (Feb) (Table 6) Capabilities
6 Jan 2026 MiMo-V2-Flash technical report 84.4 HMMT Feb. 2025 (Table 9) Capabilities
5 Jan 2026 Falcon H1R 7B technical report 64.9 HMMT25 pass@1 (Table 4) Capabilities
1 Dec 2025 DeepSeek-V3.2 technical report 92.5 HMMT Feb 2025 (Pass@1), DeepSeek-V3.2 Thinking (Table 2) Capabilities
1 Dec 2025 DeepSeek-V3.2-Speciale technical report 99.2 HMMT Feb 2025 (Pass@1), DeepSeek-V3.2-Speciale (Table 3) Capabilities
26 Nov 2025 Qwen3-VL-235B-A22B technical report 77.4 HMMT-25 (Table 6) Capabilities
21 Oct 2025 Ring-1T technical report 86.7 HMMT25 (Avg@16) (Table 1) Capabilities
29 Sep 2025 DeepSeek-V3.2-Exp technical report 83.6 HMMT 2025 (Pass@1), DeepSeek-V3.2-Exp (Table 1) Capabilities
23 Sep 2025 LongCat-Flash-Thinking technical report 83.7 HMMT-25(Mean@32) (Table 2) Capabilities
21 Jul 2025 Kimi K2 technical report 38.8 HMMT 2025 (Avg@32) (Table 3) Capabilities
15 Jul 2025 EXAONE 4.0 32B technical report 72.9 HMMT Feb 2025, n=32 (Table 3, Reasoning). Non-reasoning 21.8 (Table 4) Capabilities
30 Apr 2025 Phi-4-reasoning technical report 43.8 HMMT Feb 2025, average pass@1, MathArena scoring, std 6.2 (Table 1) Capabilities
30 Apr 2025 Phi-4-Reasoning-plus technical report 53.6 HMMT Feb 2025, average pass@1 over 64 runs, MathArena scoring (Table 1) Capabilities

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 Google DeepMind xAI
HMMT 2026 Passed 85% in Feb 2026, by GPT-5.2
GPT-5.5 98.5%
HMMT 2025 Passed 85% in Jul 2025, by Grok 4 Heavy
GPT-5 100%
HMMT 2026 Passed 85% in Feb 2026, by GPT-5.2
GPT-5.5 98.5%
Replaces HMMT February 2024 Replaced by HMMT 2026

More in Math

All 105 in Math
  • IMO-AnswerBench In use 400 olympiad problems with short verifiable answers, robustified against memorization. Scored by answer match. GLM-5.2 91%
  • 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%
  • 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%
  • 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%
  • USAMO 2026 In use Models write full proofs for the six problems of the 2026 USA Mathematical Olympiad. Judges grade each proof from 0 to 7 against rubrics. GPT-5.5 98.2%
  • ArXivMath In use Final-answer research-level math problems extracted monthly from recent arXiv papers and filtered to be self-contained. GPT-6.1 Sol 94.2%