Benchmark Almanac

SMT 2025

Gemini 3 Pro 93.4% leads since Nov 2025 Not in model cards

Models answer 53 final-answer problems from the 2025 Stanford Math Tournament. MathArena scores the average accuracy over four attempts.

MathArena ran 44 models and later marked the set deprecated. MathArena did not run a 2026 edition.

Made by Stanford Math Tournament (contest) , MathArena (evaluation) Released May 2025 Competition problems with answers All tasks public Last model added Feb 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 Paper Dataset Leaderboard stanfordmathtournament.com
40
Models scored on its own leaderboard

Gemini 3 Pro leads with 93.4% among models released by Feb 2026

Best score per model from the benchmark's own leaderboard, read on 2 Nov 2026. Where a model ran with several settings or harnesses, we keep its best. Every run is in the next chart.

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

40 models. Scroll sideways for the rest.

  1. 93.4 Gemini 3 Pro
  2. 92.9 Gemini 3 Flash
  3. 92.0 GPT-5
  4. 92.0 GPT-5.2
  5. 91.5 Step 3.5 Flash
  6. 91.0 Kimi K2 Thinking
  7. 91.0 GPT-5.1
  8. 91.0 GLM-5
  9. 90.6 GLM-4.6
  10. 90.6 Kimi K2.5
  11. 89.2 GPT-5 mini
  12. 89.2 DeepSeek-V3.2-Speciale
  13. 88.7 o4-mini
  14. 87.7 o3
  15. 87.7 gpt-oss-120b
  16. 87.7 DeepSeek-V3.2
  17. 85.8 Grok 4
  18. 85.8 Falcon H1R 7B
  19. 85.4 GPT-5 nano
  20. 84.9 Gemini 2.5 Pro
  21. 84.9 DeepSeek-V3.2-Exp
  22. 84.6 Grok 4.1 Fast
  23. 84.4 Grok 4 Fast
  24. 84.0 DeepSeek-V3.1
  25. 84.0 Claude Sonnet 4.5
  26. 83.0 DeepSeek-R1
  27. 83.0 gpt-oss-20b
  28. 82.1 GLM-4.5
  29. 79.7 K2-Think
  30. 79.7 QED-Nano
  31. 78.8 Grok-3 mini
  32. 77.4 GLM-4.5-Air
  33. 76.9 Qwen3-235B-A22B
  34. 75.5 Gemini 2.5 Flash
  35. 67.9 Qwen3-30B-A3B
  36. 67.0 DeepSeek-R1
  37. 60.9 DeepSeek-R1-Distill-Llama-70B
  38. 60.4 DeepSeek-R1-Distill-Qwen-32B
  39. 56.6 Claude 3.7 Sonnet
  40. 54.7 DeepSeek-R1-Distill-Qwen-14B
  1. 1 Gemini 3 Pro 93.4%
  2. 2 Gemini 3 Flash 92.9%
  3. 3 GPT-5 high 92%
  4. 4 GPT-5.2 high 92%
  5. 5 Step 3.5 Flash 91.5%
  6. 6 Kimi K2 Thinking 91%
  7. 7 GPT-5.1 high 91%
  8. 8 GLM-5 91%
  9. 9 GLM-4.6 90.6%
  10. 10 Kimi K2.5 thinking 90.6%
  11. 11 GPT-5 mini high 89.2%
  12. 12 DeepSeek-V3.2-Speciale 89.2%
  13. 13 o4-mini high 88.7%
  14. 14 o3 high 87.7%
  15. 15 gpt-oss-120b high 87.7%
  16. 16 DeepSeek-V3.2 thinking 87.7%
  17. 17 Grok 4 85.8%
  18. 18 Falcon H1R 7B 85.8%
  19. 19 GPT-5 nano high 85.4%
  20. 20 Gemini 2.5 Pro Mar 2025 84.9%
  21. 21 DeepSeek-V3.2-Exp thinking 84.9%
  22. 22 Grok 4.1 Fast 84.6%
  23. 23 Grok 4 Fast 84.4%
  24. 24 DeepSeek-V3.1 thinking 84%
  25. 25 Claude Sonnet 4.5 thinking 84%
  26. 26 DeepSeek-R1 May 2025 83%
  27. 27 gpt-oss-20b high 83%
  28. 28 GLM-4.5 82.1%
  29. 29 K2-Think 79.7%
  30. 30 QED-Nano 79.7%
  31. 31 Grok-3 mini high 78.8%
  32. 32 GLM-4.5-Air 77.4%
  33. 33 Qwen3-235B-A22B 76.9%
  34. 34 Gemini 2.5 Flash Sep 2025, thinking 75.5%
  35. 35 Qwen3-30B-A3B 67.9%
  36. 36 DeepSeek-R1 67%
  37. 37 DeepSeek-R1-Distill-Llama-70B 60.9%
  38. 38 DeepSeek-R1-Distill-Qwen-32B 60.4%
  39. 39 Claude 3.7 Sonnet thinking 56.6%
  40. 40 DeepSeek-R1-Distill-Qwen-14B 54.7%
#ModelScoreSettingDateSource
1 Gemini 3 Pro 93.4% Nov 2025 Official leaderboard
2 Gemini 3 Flash 92.9% Dec 2025 Official leaderboard
3 GPT-5 92% high Aug 2025 Official leaderboard
4 GPT-5.2 92% high Dec 2025 Official leaderboard
5 Step 3.5 Flash 91.5% Feb 2026 Official leaderboard
6 Kimi K2 Thinking 91% Nov 2025 Official leaderboard
7 GPT-5.1 91% high Nov 2025 Official leaderboard
8 GLM-5 91% Feb 2026 Official leaderboard
9 GLM-4.6 90.6% Sep 2025 Official leaderboard
10 Kimi K2.5 90.6% thinking Jan 2026 Official leaderboard
11 GPT-5 mini 89.2% high Aug 2025 Official leaderboard
12 DeepSeek-V3.2-Speciale 89.2% Dec 2025 Official leaderboard
13 o4-mini 88.7% high Apr 2025 Official leaderboard
14 o3 87.7% high Apr 2025 Official leaderboard
15 gpt-oss-120b 87.7% high Aug 2025 Official leaderboard
16 DeepSeek-V3.2 87.7% thinking Dec 2025 Official leaderboard
17 Grok 4 85.8% Jul 2025 Official leaderboard
18 Falcon H1R 7B 85.8% Jan 2026 Official leaderboard
19 GPT-5 nano 85.4% high Aug 2025 Official leaderboard
20 Gemini 2.5 Pro Mar 2025 84.9% Mar 2025 Official leaderboard
21 DeepSeek-V3.2-Exp 84.9% thinking Sep 2025 Official leaderboard
22 Grok 4.1 Fast 84.6% Nov 2025 Official leaderboard
23 Grok 4 Fast 84.4% Sep 2025 Official leaderboard
24 DeepSeek-V3.1 84% thinking Aug 2025 Official leaderboard
25 Claude Sonnet 4.5 84% thinking Sep 2025 Official leaderboard
26 DeepSeek-R1 May 2025 83% May 2025 Official leaderboard
27 gpt-oss-20b 83% high Aug 2025 Official leaderboard
28 GLM-4.5 82.1% Aug 2025 Official leaderboard
29 K2-Think 79.7% Sep 2025 Official leaderboard
30 QED-Nano 79.7% Feb 2026 Official leaderboard
31 Grok-3 mini 78.8% high Apr 2025 Official leaderboard
32 GLM-4.5-Air 77.4% Jul 2025 Official leaderboard
33 Qwen3-235B-A22B 76.9% Apr 2025 Official leaderboard
34 Gemini 2.5 Flash Sep 2025 75.5% thinking Sep 2025 Official leaderboard
35 Qwen3-30B-A3B 67.9% Apr 2025 Official leaderboard
36 DeepSeek-R1 67% Jan 2025 Official leaderboard
37 DeepSeek-R1-Distill-Llama-70B 60.9% Jan 2025 Official leaderboard
38 DeepSeek-R1-Distill-Qwen-32B 60.4% Jan 2025 Official leaderboard
39 Claude 3.7 Sonnet 56.6% thinking Feb 2025 Official leaderboard
40 DeepSeek-R1-Distill-Qwen-14B 54.7% Jan 2025 Official leaderboard

The best score, 93.4%, costs $0.17 a task

Runs on the cost frontier score higher than any run that's cheaper. DeepSeek-V3.2-Speciale comes within 4.3 points of the best for $0.009 a task, 19 times cheaper. The costs are the leaderboard's own estimates, so compare them only within this benchmark.

xAI Google DeepMind 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 Cost per task, log scale Gemini 3 Pro Gemini 3 Flash Step 3.5 Flash DeepSeek-V3.2, thinking Grok-3 mini, low 0% 20% 40% 60% 80% 100% $0.001 $0.01 $0.10 $1.00 Cost per task, log scale Gemini 3 Pro Step 3.5 Flash
ModelEffortHarnessScoreCost per taskOutput tokens
Gemini 3 Pro 93.4% $0.17 13,898
Gemini 3 Flash 92.9% $0.06 18,434
GPT-5 high 92% $0.12 11,731
GPT-5.2 high 92% $0.13 9,214
Step 3.5 Flash 91.5% $0.01 39,239
Kimi K2 Thinking 91% $0.05 21,526
GPT-5.1 high 91% $0.16 15,797
GLM-5 91% $0.08 24,104
GLM-4.6 90.6% $0.03 15,207
Kimi K2.5 thinking 90.6% $0.07 23,585
GPT-5 mini high 89.2% $0.02 12,000
DeepSeek-V3.2-Speciale 89.2% $0.009 20,448
o4-mini high 88.7% $0.04 10,276
o3 high 87.7% $0.07 8,524
gpt-oss-120b high 87.7% $0.01 18,349
DeepSeek-V3.2 thinking 87.7% $0.006 14,361
Grok 4 85.8% $0.18 12,194
Falcon H1R 7B 85.8% $0.004 18,229
GPT-5 nano high 85.4% $0.02 38,582
Gemini 2.5 Pro Mar 2025 84.9% $0.19 18,603
DeepSeek-V3.2-Exp thinking 84.9% $0.006 14,122
Grok 4.1 Fast 84.6% $0.01 21,340
Grok 4 Fast 84.4% $0.005 9,219
DeepSeek-V3.1 thinking 84% $0.03 15,144
Claude Sonnet 4.5 thinking 84% $0.24 15,966
DeepSeek-R1 May 2025 83% $0.04 20,491
gpt-oss-20b high 83% $0.007 35,182
GLM-4.5 82.1% $0.05 21,383
o4-mini medium 79.7% $0.02 4,590
K2-Think 79.7%
QED-Nano 79.7% 29,553
Grok-3 mini high 78.8% $0.009 17,682
GLM-4.5-Air 77.4% $0.03 23,047
Qwen3-235B-A22B 76.9% $0.008 13,024
Gemini 2.5 Flash Sep 2025 thinking 75.5% $0.08 21,599
o4-mini low 68.9% $0.009 1,997
Qwen3-30B-A3B 67.9% $0.004 14,870
DeepSeek-R1 67% $0.02 10,323
Grok-3 mini low 63.7% $0.003 6,086
DeepSeek-R1-Distill-Llama-70B 60.9% $0.006 10,049
DeepSeek-R1-Distill-Qwen-32B 60.4% $0.007 11,380
Claude 3.7 Sonnet thinking 56.6% $0.34 22,813
DeepSeek-R1-Distill-Qwen-14B 54.7% $0.004 12,399

3 models have held number one since its release in May 2025

Lab claims count from the day the card came out, independent runs from the model’s release.

OpenAI Google DeepMind
Own leaderboard Passed 85% in May 2025, by o4-mini
Gemini 3 Pro 93.4%

The best score passed 85% in Apr 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 Best so far
0% 20% 40% 60% 80% 100% 2025 2026 Near the ceiling (85%) Gemini 3 Pro GPT-5 o4-mini DeepSeek-R1 0% 20% 40% 60% 80% 100% 2025 2026 Near the ceiling (85%) Gemini 3 Pro Gemini 2.5 Pro

No lab model card reports it yet

We read the headline results of 479 launch posts from 51 labs, and none of them list this benchmark.

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%
  • 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%
  • 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%