Benchmark Almanac

MathArena Apex Shortlist

GPT-5.5 98.4% leads since April 2026
GPT-5.5 98.4% leads since April 2026
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.

MathArena picked the Apex Shortlist and the 12-problem Apex set together in August 2025.

Released
August 2025
Kind
Competition problems with answers
Level
Competition
Scored
Checked automatically
Answers
Short answer
Tasks
All public
Strongest model tested
GPT-5.5 #20 on the capability index
Last in a launch post
October 2026
Last model added
October 2026
Scores refreshed
November 12th, 2026
38
Models scored on its own leaderboard
4
Launch posts list it, from 2 labs
2
System cards report it, from 1 lab

GPT-5.5 leads with 98.4% among models released by July 2026

Best score per model from the benchmark's own leaderboard, read on November 12th, 2026. Where a model ran with several settings or harnesses, we keep its best. Every run is in the next chart.

GPT-5.5 leads with 98.4%, 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 tested a newer one. Among the 20 highest-ranked models on the capability index, these have no score yet:

38 models. Scroll sideways for the rest.

  1. 98.4 GPT-5.5
  2. 92.0 Gemini 3.1 Pro
  3. 92.0 Kimi K3
  4. 90.4 Claude Opus 4.8
  5. 89.4 DeepSeek-V4-Flash
  6. 87.8 DeepSeek-V4-Pro
  7. 86.7 Claude Opus 4.6
  8. 82.5 Gemini 3.5 Flash
  9. 81.4 GPT-5.4
  10. 79.3 GPT-5.2
  11. 77.1 Kimi K2.6
  12. 76.6 Step 3.7 Flash
  13. 71.8 GLM-5.1
  14. 71.3 Gemini 3.6 Flash
  15. 70.7 Step 3.5 Flash
  16. 69.7 DeepSeek-V3.2-Speciale
  17. 68.6 Gemini 3 Flash
  18. 68.6 GLM-5
  19. 68.1 GLM-5.2
  20. 66.5 Gemini 3 Pro
  21. 63.8 Claude Opus 4.7
  22. 60.1 Qwen3.5 397B-A17B
  23. 57.9 Grok 4.1 Fast
  24. 57.5 Grok 4
  25. 57.5 Kimi K2.5
  26. 57.5 Nemotron 3 Super
  27. 55.9 GPT-5.1
  28. 55.3 Grok 4 Fast
  29. 52.1 Qwen3.5-27B
  30. 50.5 DeepSeek-V3.2
  31. 47.3 Kimi K2 Thinking
  32. 45.2 gpt-oss-120b
  33. 44.7 Qwen3.5-35B-A3B
  34. 39.9 GPT-5 mini
  35. 29.8 Qwen3.5-9B
  36. 23.4 Qwen3-30B-A3B-Thinking
  37. 22.3 QED-Nano
  38. 16.5 Qwen3-4B-Thinking-2507
  1. 1 GPT-5.5 xhigh 98.4%
  2. 2 Gemini 3.1 Pro high 92%
  3. 3 Kimi K3 max 92%
  4. 4 Claude Opus 4.8 max 90.4%
  5. 5 DeepSeek-V4-Flash max 89.4%
  6. 6 DeepSeek-V4-Pro max 87.8%
  7. 7 Claude Opus 4.6 high 86.7%
  8. 8 Gemini 3.5 Flash high 82.5%
  9. 9 GPT-5.4 xhigh 81.4%
  10. 10 GPT-5.2 high 79.3%
  11. 11 Kimi K2.6 high 77.1%
  12. 12 Step 3.7 Flash high 76.6%
  13. 13 GLM-5.1 71.8%
  14. 14 Gemini 3.6 Flash 71.3%
  15. 15 Step 3.5 Flash 70.7%
  16. 16 DeepSeek-V3.2-Speciale 69.7%
  17. 17 Gemini 3 Flash high 68.6%
  18. 18 GLM-5 68.6%
  19. 19 GLM-5.2 max 68.1%
  20. 20 Gemini 3 Pro 66.5%
  21. 21 Claude Opus 4.7 xhigh 63.8%
  22. 22 Qwen3.5 397B-A17B 60.1%
  23. 23 Grok 4.1 Fast reasoning 57.9%
  24. 24 Grok 4 57.5%
  25. 25 Kimi K2.5 high 57.5%
  26. 26 Nemotron 3 Super 57.5%
  27. 27 GPT-5.1 high 55.9%
  28. 28 Grok 4 Fast 55.3%
  29. 29 Qwen3.5-27B 52.1%
  30. 30 DeepSeek-V3.2 thinking 50.5%
  31. 31 Kimi K2 Thinking 47.3%
  32. 32 gpt-oss-120b high 45.2%
  33. 33 Qwen3.5-35B-A3B 44.7%
  34. 34 GPT-5 mini high 39.9%
  35. 35 Qwen3.5-9B 29.8%
  36. 36 Qwen3-30B-A3B-Thinking Jul 2025 23.4%
  37. 37 QED-Nano 22.3%
  38. 38 Qwen3-4B-Thinking-2507 16.5%

Source: Official MathArena Apex Shortlist leaderboard

40 models. Scroll sideways for the rest.

  1. 98.4 GPT-5.5
  2. 92.0 Gemini 3.1 Pro
  3. 92.0 Kimi K3
  4. 90.4 Claude Opus 4.8
  5. 90.2 DeepSeek-V4-Pro
  6. 89.4 DeepSeek-V4-Flash
  7. 88.4 Nemotron 3.5 Super VL
  8. 86.7 Claude Opus 4.6
  9. 82.5 Gemini 3.5 Flash
  10. 81.4 GPT-5.4
  11. 79.3 GPT-5.2
  12. 77.1 Kimi K2.6
  13. 76.6 Step 3.7 Flash
  14. 74.9 Nemotron 3 Ultra
  15. 71.8 GLM-5.1
  16. 71.3 Gemini 3.6 Flash
  17. 70.7 Step 3.5 Flash
  18. 69.7 DeepSeek-V3.2-Speciale
  19. 68.6 Gemini 3 Flash
  20. 68.6 GLM-5
  21. 68.1 GLM-5.2
  22. 66.5 Gemini 3 Pro
  23. 63.8 Claude Opus 4.7
  24. 60.1 Qwen3.5 397B-A17B
  25. 57.9 Grok 4.1 Fast
  26. 57.5 Grok 4
  27. 57.5 Kimi K2.5
  28. 57.5 Nemotron 3 Super
  29. 55.9 GPT-5.1
  30. 55.3 Grok 4 Fast
  31. 52.1 Qwen3.5-27B
  32. 50.5 DeepSeek-V3.2
  33. 47.3 Kimi K2 Thinking
  34. 45.2 gpt-oss-120b
  35. 44.7 Qwen3.5-35B-A3B
  36. 39.9 GPT-5 mini
  37. 29.8 Qwen3.5-9B
  38. 23.4 Qwen3-30B-A3B-Thinking
  39. 22.3 QED-Nano
  40. 16.5 Qwen3-4B-Thinking-2507
  1. 1 GPT-5.5 xhigh 98.4%
  2. 2 Gemini 3.1 Pro high 92%
  3. 3 Kimi K3 max 92%
  4. 4 Claude Opus 4.8 max 90.4%
  5. 5 DeepSeek-V4-Pro 90.2%
  6. 6 DeepSeek-V4-Flash max 89.4%
  7. 7 Nemotron 3.5 Super VL 88.4%
  8. 8 Claude Opus 4.6 high 86.7%
  9. 9 Gemini 3.5 Flash high 82.5%
  10. 10 GPT-5.4 xhigh 81.4%
  11. 11 GPT-5.2 high 79.3%
  12. 12 Kimi K2.6 high 77.1%
  13. 13 Step 3.7 Flash high 76.6%
  14. 14 Nemotron 3 Ultra 74.9%
  15. 15 GLM-5.1 71.8%
  16. 16 Gemini 3.6 Flash 71.3%
  17. 17 Step 3.5 Flash 70.7%
  18. 18 DeepSeek-V3.2-Speciale 69.7%
  19. 19 Gemini 3 Flash high 68.6%
  20. 20 GLM-5 68.6%
  21. 21 GLM-5.2 max 68.1%
  22. 22 Gemini 3 Pro 66.5%
  23. 23 Claude Opus 4.7 xhigh 63.8%
  24. 24 Qwen3.5 397B-A17B 60.1%
  25. 25 Grok 4.1 Fast reasoning 57.9%
  26. 26 Grok 4 57.5%
  27. 27 Kimi K2.5 high 57.5%
  28. 28 Nemotron 3 Super 57.5%
  29. 29 GPT-5.1 high 55.9%
  30. 30 Grok 4 Fast 55.3%
  31. 31 Qwen3.5-27B 52.1%
  32. 32 DeepSeek-V3.2 thinking 50.5%
  33. 33 Kimi K2 Thinking 47.3%
  34. 34 gpt-oss-120b high 45.2%
  35. 35 Qwen3.5-35B-A3B 44.7%
  36. 36 GPT-5 mini high 39.9%
  37. 37 Qwen3.5-9B 29.8%
  38. 38 Qwen3-30B-A3B-Thinking Jul 2025 23.4%
  39. 39 QED-Nano 22.3%
  40. 40 Qwen3-4B-Thinking-2507 16.5%

Sources: Official MathArena Apex Shortlist leaderboard and labs’ model cards

#ModelScoreSettingDateSource
1 GPT-5.5 98.4% xhigh Apr 2026 Official leaderboard
2 Gemini 3.1 Pro 92% high Feb 2026 Official leaderboard
3 Kimi K3 92% max Jul 2026 Official leaderboard
4 Claude Opus 4.8 90.4% max May 2026 Official leaderboard
5 DeepSeek-V4-Flash 89.4% max Apr 2026 Official leaderboard
6 DeepSeek-V4-Pro 87.8% max Apr 2026 Official leaderboard
7 Claude Opus 4.6 86.7% high Feb 2026 Official leaderboard
8 Gemini 3.5 Flash 82.5% high May 2026 Official leaderboard
9 GPT-5.4 81.4% xhigh Mar 2026 Official leaderboard
10 GPT-5.2 79.3% high Dec 2025 Official leaderboard
11 Kimi K2.6 77.1% high Apr 2026 Official leaderboard
12 Step 3.7 Flash 76.6% high May 2026 Official leaderboard
13 GLM-5.1 71.8% Apr 2026 Official leaderboard
14 Gemini 3.6 Flash 71.3% Jul 2026 Official leaderboard
15 Step 3.5 Flash 70.7% Feb 2026 Official leaderboard
16 DeepSeek-V3.2-Speciale 69.7% Dec 2025 Official leaderboard
17 Gemini 3 Flash 68.6% high Dec 2025 Official leaderboard
18 GLM-5 68.6% Feb 2026 Official leaderboard
19 GLM-5.2 68.1% max Jun 2026 Official leaderboard
20 Gemini 3 Pro 66.5% Nov 2025 Official leaderboard
21 Claude Opus 4.7 63.8% xhigh Apr 2026 Official leaderboard
22 Qwen3.5 397B-A17B 60.1% Feb 2026 Official leaderboard
23 Grok 4.1 Fast 57.9% reasoning Nov 2025 Official leaderboard
24 Grok 4 57.5% Jul 2025 Official leaderboard
25 Kimi K2.5 57.5% high Jan 2026 Official leaderboard
26 Nemotron 3 Super 57.5% Mar 2026 Official leaderboard
27 GPT-5.1 55.9% high Nov 2025 Official leaderboard
28 Grok 4 Fast 55.3% Sep 2025 Official leaderboard
29 Qwen3.5-27B 52.1% Feb 2026 Official leaderboard
30 DeepSeek-V3.2 50.5% thinking Dec 2025 Official leaderboard
31 Kimi K2 Thinking 47.3% Nov 2025 Official leaderboard
32 gpt-oss-120b 45.2% high Aug 2025 Official leaderboard
33 Qwen3.5-35B-A3B 44.7% Feb 2026 Official leaderboard
34 GPT-5 mini 39.9% high Aug 2025 Official leaderboard
35 Qwen3.5-9B 29.8% Feb 2026 Official leaderboard
36 Qwen3-30B-A3B-Thinking Jul 2025 23.4% Jul 2025 Official leaderboard
37 QED-Nano 22.3% Feb 2026 Official leaderboard
38 Qwen3-4B-Thinking-2507 16.5% Aug 2025 Official leaderboard
#ModelScoreSettingDateSource
1 GPT-5.5 98.4% xhigh Apr 2026 Official leaderboard
2 Gemini 3.1 Pro 92% high Feb 2026 Official leaderboard
3 Kimi K3 92% max Jul 2026 Official leaderboard
4 Claude Opus 4.8 90.4% max May 2026 Official leaderboard
5 DeepSeek-V4-Pro 90.2% Apr 2026 Lab model card
6 DeepSeek-V4-Flash 89.4% max Apr 2026 Official leaderboard
7 Nemotron 3.5 Super VL 88.4% Oct 2026 Lab model card
8 Claude Opus 4.6 86.7% high Feb 2026 Official leaderboard
9 Gemini 3.5 Flash 82.5% high May 2026 Official leaderboard
10 GPT-5.4 81.4% xhigh Mar 2026 Official leaderboard
11 GPT-5.2 79.3% high Dec 2025 Official leaderboard
12 Kimi K2.6 77.1% high Apr 2026 Official leaderboard
13 Step 3.7 Flash 76.6% high May 2026 Official leaderboard
14 Nemotron 3 Ultra 74.9% Jun 2026 Lab model card
15 GLM-5.1 71.8% Apr 2026 Official leaderboard
16 Gemini 3.6 Flash 71.3% Jul 2026 Official leaderboard
17 Step 3.5 Flash 70.7% Feb 2026 Official leaderboard
18 DeepSeek-V3.2-Speciale 69.7% Dec 2025 Official leaderboard
19 Gemini 3 Flash 68.6% high Dec 2025 Official leaderboard
20 GLM-5 68.6% Feb 2026 Official leaderboard
21 GLM-5.2 68.1% max Jun 2026 Official leaderboard
22 Gemini 3 Pro 66.5% Nov 2025 Official leaderboard
23 Claude Opus 4.7 63.8% xhigh Apr 2026 Official leaderboard
24 Qwen3.5 397B-A17B 60.1% Feb 2026 Official leaderboard
25 Grok 4.1 Fast 57.9% reasoning Nov 2025 Official leaderboard
26 Grok 4 57.5% Jul 2025 Official leaderboard
27 Kimi K2.5 57.5% high Jan 2026 Official leaderboard
28 Nemotron 3 Super 57.5% Mar 2026 Official leaderboard
29 GPT-5.1 55.9% high Nov 2025 Official leaderboard
30 Grok 4 Fast 55.3% Sep 2025 Official leaderboard
31 Qwen3.5-27B 52.1% Feb 2026 Official leaderboard
32 DeepSeek-V3.2 50.5% thinking Dec 2025 Official leaderboard
33 Kimi K2 Thinking 47.3% Nov 2025 Official leaderboard
34 gpt-oss-120b 45.2% high Aug 2025 Official leaderboard
35 Qwen3.5-35B-A3B 44.7% Feb 2026 Official leaderboard
36 GPT-5 mini 39.9% high Aug 2025 Official leaderboard
37 Qwen3.5-9B 29.8% Feb 2026 Official leaderboard
38 Qwen3-30B-A3B-Thinking Jul 2025 23.4% Jul 2025 Official leaderboard
39 QED-Nano 22.3% Feb 2026 Official leaderboard
40 Qwen3-4B-Thinking-2507 16.5% Aug 2025 Official leaderboard

Labs report scores for 4 models in their cards. Switch the scores above to rank them together.

The best score, 98.4%, costs $0.77 a task

Runs on the cost frontier score higher than any run that's cheaper. The costs are the leaderboard's own estimates, so compare them only within this benchmark.

xAI DeepSeek StepFun Google DeepMind Qwen OpenAI 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-5.5, xhigh Gemini 3.1 Pro, high DeepSeek-V4-Flash, max Step 3.5 Flash Grok 4.1 Fast, reasoning Qwen3.5-9B 0% 20% 40% 60% 80% 100% $0.001 $0.01 $0.10 $1.00 $10 Cost per task, log scale GPT-5.5, xhigh Gemini 3.1 Pro, high

Source: Official MathArena Apex Shortlist leaderboard

ModelEffortHarnessScoreCost per taskOutput tokens
GPT-5.5 xhigh 98.4% $0.77 25,639
Gemini 3.1 Pro high 92% $0.37 30,656
Kimi K3 max 92% $0.66 44,231
Claude Opus 4.8 max 90.4% $3.14 125,466
DeepSeek-V4-Flash max 89.4% $0.04 142,705
DeepSeek-V4-Pro max 87.8% $0.34 98,925
Claude Opus 4.6 high 86.7% $1.80 71,911
Gemini 3.5 Flash high 82.5% $0.26 29,307
GPT-5.4 xhigh 81.4% $0.53 33,701
GPT-5.2 high 79.3% $0.76 54,048
Kimi K2.6 high 77.1% $0.27 68,577
Step 3.7 Flash high 76.6% $0.06 207,503
GLM-5.1 71.8% $0.24 73,757
Gemini 3.6 Flash 71.3% $0.19 25,211
Step 3.5 Flash 70.7% $0.04 129,842
DeepSeek-V3.2-Speciale 69.7% $0.03 66,236
Gemini 3 Flash high 68.6% $0.09 31,434
GLM-5 68.6% $0.22 68,201
GLM-5.2 max 68.1% $0.16 38,771
Gemini 3 Pro 66.5% $0.27 22,821
Claude Opus 4.7 xhigh 63.8% $1.83 72,956
Qwen3.5 397B-A17B 60.1% $0.14 39,424
Grok 4.1 Fast reasoning 57.9% $0.01 26,925
Grok 4 57.5% $0.52 34,614
Kimi K2.5 high 57.5% $0.16 52,032
Nemotron 3 Super 57.5% 90,579
GPT-5.1 high 55.9% $0.57 57,017
Grok 4 Fast 55.3% $0.01 24,231
Qwen3.5-27B 52.1% $0.15 60,409
DeepSeek-V3.2 thinking 50.5% $0.02 38,748
Kimi K2 Thinking 47.3% $0.14 57,238
gpt-oss-120b high 45.2% $0.03 42,972
Qwen3.5-35B-A3B 44.7% $0.09 65,307
GPT-5 mini high 39.9% $0.07 33,794
Qwen3.5-9B 29.8% $0.009 62,059
Qwen3-30B-A3B-Thinking Jul 2025 23.4% $0.01 32,111
QED-Nano 22.3% 97,948
Qwen3-4B-Thinking-2507 16.5% $0.01 32,631

7 models have held number one on the official leaderboard since its release in August 2025 7 models have held number one with claims included since its release in August 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.

xAI Google DeepMind DeepSeek OpenAI Anthropic
Own leaderboard Passed 85% in Feb 2026, by Claude Opus 4.6
GPT-5.5 98.4%
Lab claims Passed 85% in Apr 2026, by DeepSeek-V4-Pro
DeepSeek-V4-Pro 90.2%

Sources: Official MathArena Apex Shortlist leaderboard and labs’ model cards

Claude Opus 4.6 on the official leaderboard passed 85% in February 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% 2026 Near the ceiling (85%) GPT-5.5 Gemini 3.1 Pro Claude Opus 4.6 GPT-5.2 DeepSeek-V3.2-Speciale Grok 4 0% 20% 40% 60% 80% 100% 2026 Near the ceiling (85%) GPT-5.5 Gemini 3.1 Pro

Sources: Official MathArena Apex Shortlist leaderboard and labs’ model cards

One lab 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.

DeepSeek
NVIDIA
DateLaunch postScoreAs the post writes it
Oct 15, 2026 Nemotron 3.5 Super VL 88.4 APEX shortlist
Jun 4, 2026 Nemotron 3 Ultra 74.9 Apex-Shortlist (no tools); with tools 84.8
Apr 24, 2026 DeepSeek-V4-Flash 85.7 Apex Shortlist (Pass@1), V4-Flash Max
Apr 24, 2026 DeepSeek-V4-Pro 90.2 Apex Shortlist (Pass@1)

2 system cards report it, from 1 lab

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

DateDocumentScoreAs the document writes itSection
Oct 15, 2026 Nemotron 3.5 Super VL technical report 88.4 Apex-Shortlist, no tools, mean of 16 (Table 10) Capabilities
Jun 9, 2026 Nemotron 3 Ultra technical report 74.9 Apex-Shortlist (no tools), 84.8 with tools (Table 10) Capabilities

The lab's number comes out ahead in 1 of 2 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. Every benchmark.

Model Lab says Board says Gap
DeepSeek-V4-Flash 85.7% Apex Shortlist (Pass@1), V4-Flash Max 89.4% Leaderboard, max −3.7 pts
DeepSeek-V4-Pro 90.2% Apex Shortlist (Pass@1) 87.8% Leaderboard, max +2.4 pts

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%
  • 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%
  • 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%