Benchmark Almanac

AIME 2026

GPT-5.5 100% leads since Apr 2026 In use

Models solve the 30 integer-answer problems from the 2026 AIME I and II math competitions. Scored by exact answer match.

An exam, not a dedicated benchmark. Top models score above 90%.

Made by Mathematical Association of America (exam) , MathArena (evaluation) Released Feb 2026 Competition problems with answers All tasks public Last in a launch post Oct 2026 Last model added Oct 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 matharena.ai
31
Models scored on its own leaderboard
31
Launch posts list it, from 17 labs
20
System cards report it, from 13 labs

2 models share the top score of 100% among models released by Jul 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 Jul 2026, and nobody has tested a newer one. Among the 20 highest-ranked models on the capability index, these have no score yet:

31 models. Scroll sideways for the rest.

  1. 100.0 GPT-5.5
  2. 100.0 Claude Opus 4.8
  3. 99.2 GPT-5.4
  4. 98.3 GPT-5.2
  5. 98.3 Gemini 3.1 Pro
  6. 96.7 Gemini 3 Flash
  7. 96.7 Step 3.5 Flash
  8. 96.7 Claude Opus 4.6
  9. 96.7 GLM-5
  10. 96.7 DeepSeek-V4-Pro
  11. 96.7 Kimi K3
  12. 96.7 Gemini 3.6 Flash
  13. 95.8 Kimi K2.5
  14. 95.8 GLM-5.1
  15. 95.8 Claude Opus 4.7
  16. 95.8 Kimi K2.6
  17. 95.8 DeepSeek-V4-Flash
  18. 95.0 Gemini 3.5 Flash
  19. 95.0 Step 3.7 Flash
  20. 94.2 Grok 4.1 Fast
  21. 94.2 DeepSeek-V3.2
  22. 94.2 Qwen3.5 397B-A17B
  23. 93.3 Qwen3.5-35B-A3B
  24. 92.5 Qwen3.5-9B
  25. 91.7 Gemini 3 Pro
  26. 91.7 Qwen3.5-27B
  27. 91.7 Nemotron 3 Super
  28. 90.0 GLM-5.2
  29. 88.3 Qwen3-30B-A3B-Thinking
  30. 82.5 Qwen3-4B-Thinking-2507
  31. 82.5 QED-Nano
  1. 1 GPT-5.5 xhigh 100%
  2. 2 Claude Opus 4.8 max 100%
  3. 3 GPT-5.4 xhigh 99.2%
  4. 4 GPT-5.2 high 98.3%
  5. 5 Gemini 3.1 Pro 98.3%
  6. 6 Gemini 3 Flash 96.7%
  7. 7 Step 3.5 Flash 96.7%
  8. 8 Claude Opus 4.6 high 96.7%
  9. 9 GLM-5 96.7%
  10. 10 DeepSeek-V4-Pro max 96.7%
  11. 11 Kimi K3 thinking 96.7%
  12. 12 Gemini 3.6 Flash 96.7%
  13. 13 Kimi K2.5 thinking 95.8%
  14. 14 GLM-5.1 95.8%
  15. 15 Claude Opus 4.7 xhigh 95.8%
  16. 16 Kimi K2.6 thinking 95.8%
  17. 17 DeepSeek-V4-Flash max 95.8%
  18. 18 Gemini 3.5 Flash 95%
  19. 19 Step 3.7 Flash 95%
  20. 20 Grok 4.1 Fast 94.2%
  21. 21 DeepSeek-V3.2 thinking 94.2%
  22. 22 Qwen3.5 397B-A17B 94.2%
  23. 23 Qwen3.5-35B-A3B 93.3%
  24. 24 Qwen3.5-9B 92.5%
  25. 25 Gemini 3 Pro 91.7%
  26. 26 Qwen3.5-27B 91.7%
  27. 27 Nemotron 3 Super 91.7%
  28. 28 GLM-5.2 90%
  29. 29 Qwen3-30B-A3B-Thinking Jul 2025 88.3%
  30. 30 Qwen3-4B-Thinking-2507 82.5%
  31. 31 QED-Nano 82.5%
#ModelScoreSettingDateSource
1 GPT-5.5 100% xhigh Apr 2026 Official leaderboard
2 Claude Opus 4.8 100% max May 2026 Official leaderboard
3 GPT-5.4 99.2% xhigh Mar 2026 Official leaderboard
4 GPT-5.2 98.3% high Dec 2025 Official leaderboard
5 Gemini 3.1 Pro 98.3% Feb 2026 Official leaderboard
6 Gemini 3 Flash 96.7% Dec 2025 Official leaderboard
7 Step 3.5 Flash 96.7% Feb 2026 Official leaderboard
8 Claude Opus 4.6 96.7% high Feb 2026 Official leaderboard
9 GLM-5 96.7% Feb 2026 Official leaderboard
10 DeepSeek-V4-Pro 96.7% max Apr 2026 Official leaderboard
11 Kimi K3 96.7% thinking Jul 2026 Official leaderboard
12 Gemini 3.6 Flash 96.7% Jul 2026 Official leaderboard
13 Kimi K2.5 95.8% thinking Jan 2026 Official leaderboard
14 GLM-5.1 95.8% Apr 2026 Official leaderboard
15 Claude Opus 4.7 95.8% xhigh Apr 2026 Official leaderboard
16 Kimi K2.6 95.8% thinking Apr 2026 Official leaderboard
17 DeepSeek-V4-Flash 95.8% max Apr 2026 Official leaderboard
18 Gemini 3.5 Flash 95% May 2026 Official leaderboard
19 Step 3.7 Flash 95% May 2026 Official leaderboard
20 Grok 4.1 Fast 94.2% Nov 2025 Official leaderboard
21 DeepSeek-V3.2 94.2% thinking Dec 2025 Official leaderboard
22 Qwen3.5 397B-A17B 94.2% Feb 2026 Official leaderboard
23 Qwen3.5-35B-A3B 93.3% Feb 2026 Official leaderboard
24 Qwen3.5-9B 92.5% Feb 2026 Official leaderboard
25 Gemini 3 Pro 91.7% Nov 2025 Official leaderboard
26 Qwen3.5-27B 91.7% Feb 2026 Official leaderboard
27 Nemotron 3 Super 91.7% Mar 2026 Official leaderboard
28 GLM-5.2 90% Jun 2026 Official leaderboard
29 Qwen3-30B-A3B-Thinking Jul 2025 88.3% Jul 2025 Official leaderboard
30 Qwen3-4B-Thinking-2507 82.5% Aug 2025 Official leaderboard
31 QED-Nano 82.5% Feb 2026 Official leaderboard

Labs report scores for 34 models in their cards. We plot every source in the charts below.

The best score, 100%, costs $0.16 a task

Runs on the cost frontier score higher than any run that's cheaper. DeepSeek-V4-Flash at max comes within 4.2 points of the best for $0.009 a task, 18 times cheaper. The costs are the leaderboard's own estimates, so compare them only within this benchmark.

xAI OpenAI StepFun DeepSeek 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 GPT-5.5, xhigh Step 3.5 Flash Grok 4.1 Fast 0% 20% 40% 60% 80% 100% $0.001 $0.01 $0.10 $1.00 Cost per task, log scale GPT-5.5, xhigh Step 3.5 Flash
ModelEffortHarnessScoreCost per taskOutput tokens
GPT-5.5 xhigh 100% $0.16 5,219
Claude Opus 4.8 max 100% $0.54 21,360
GPT-5.4 xhigh 99.2% $0.16 10,743
GPT-5.2 high 98.3% $0.12 8,403
Gemini 3.1 Pro 98.3% $0.17 14,364
Gemini 3 Flash 96.7% $0.06 20,504
Step 3.5 Flash 96.7% $0.01 42,072
Claude Opus 4.6 high 96.7% $0.33 13,329
GLM-5 96.7% $0.07 23,541
DeepSeek-V4-Pro max 96.7% $0.08 23,567
Kimi K3 thinking 96.7% $0.11 7,031
Gemini 3.6 Flash 96.7% $0.08 10,951
Kimi K2.5 thinking 95.8% $0.07 22,849
GLM-5.1 95.8% $0.09 26,546
Claude Opus 4.7 xhigh 95.8% $0.27 10,728
Kimi K2.6 thinking 95.8% $0.09 22,722
DeepSeek-V4-Flash max 95.8% $0.009 33,588
Gemini 3.5 Flash 95% $0.13 13,992
Step 3.7 Flash 95% $0.01 50,365
Grok 4.1 Fast 94.2% $0.005 9,618
DeepSeek-V3.2 thinking 94.2% $0.006 14,854
Qwen3.5 397B-A17B 94.2% $0.07 20,797
Qwen3.5-35B-A3B 93.3% $0.04 31,856
Qwen3.5-9B 92.5% $0.005 31,184
Gemini 3 Pro 91.7% $0.18 14,712
Qwen3.5-27B 91.7% $0.05 29,363
Nemotron 3 Super 91.7% 24,114
GLM-5.2 90% $0.08 18,424
Qwen3-30B-A3B-Thinking Jul 2025 88.3% $0.005 17,939
Qwen3-4B-Thinking-2507 82.5% $0.006 21,206
QED-Nano 82.5% 35,191

3 models have held number one since its release in Feb 2026

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 Ant Group Qwen Moonshot AI Baidu DeepSeek
Own leaderboard Passed 85% in Feb 2026, by GPT-5.2
GPT-5.5 100%
Lab claims Passed 85% in Feb 2026, by Ring-2.5-1T
DeepSeek V4.1 Flash 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% 2026 2027 Near the ceiling (85%) GPT-5.5 GPT-5.2 Grok 4.1 Fast Qwen3-30B-A3B-Thinking 0% 20% 40% 60% 80% 100% 2026 2027 Near the ceiling (85%) GPT-5.5 Gemini 3 Pro

9 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.

Ant Group
Z.ai
ByteDance Seed
Qwen
NVIDIA
Upstage
Google DeepMind
LG AI Research
Moonshot AI
Baidu
Kuaishou
Liquid AI
Microsoft
Thinking Machines
Meta
Reflection
iFlytek
DateLaunch postScoreAs the post writes it
22 Oct 2026 Spark-X2.6-4B 91.2 AIME 2026
5 Oct 2026 Beam 1.0 97.8 AIME 2026
11 Aug 2026 Solar Pro 4 95.3 AIME 2026* (in-house)
10 Aug 2026 Muse Glimmer 94.7 AIME 2026
31 Jul 2026 K-EXAONE 2.0 92.3 AIME 2026
30 Jul 2026 Inkling-Small 95.5 AIME 2026
22 Jul 2026 Solar Open 2 95.7 AIME2026 (MathArena)
15 Jul 2026 Inkling 97.1 AIME 2026
16 Jun 2026 GLM-5.2 99.2 AIME 2026
10 Jun 2026 DiffusionGemma 26B A4B 69.1 AIME 2026 no tools
2 Jun 2026 MAI-Code-1-Flash 92.5 AIME 2026
2 Jun 2026 MAI-Thinking-1 94.5 AIME 2026
28 May 2026 LFM2.5-8B-A1B 50 AIME26
25 May 2026 Keye-VL-2.0-30B-A3B 93.3 AIME2026
14 May 2026 Ring-2.6-1T 95.8 AIME 26 (xhigh)
9 May 2026 ERNIE 5.1 99.6 Best of AIME26 w/tools (avg@32) 99.6 / AIME26 (avg@32) 95.0
29 Apr 2026 Ling-2.6-1T 87.4 AIME26 (non-reasoning, chart)
22 Apr 2026 Qwen3.6 27B 94.1 AIME26 (full AIME 2026 I and II)
20 Apr 2026 Kimi K2.6 96.4 AIME 2026
15 Apr 2026 Qwen 3.6 35B-A3B 92.7 AIME26 (full AIME 2026 I and II)
9 Apr 2026 EXAONE 4.5 33B 92.6 AIME 2026
7 Apr 2026 GLM-5.1 95.3 AIME 2026
2 Apr 2026 Gemma 4 26B A4B 88.3 AIME 2026 no tools
2 Apr 2026 Qwen 3.6 Plus 95.3 AIME26
2 Apr 2026 Gemma 4 31B IT 89.2 AIME 2026 no tools
24 Mar 2026 Solar Pro 3 (March 2026 update) 83 AIME'26
19 Mar 2026 Nemotron-Cascade 2 30B-A3B 90.9 AIME 2026 (TIR 95.0)
16 Feb 2026 Qwen3.5 397B-A17B 91.3 AIME26
14 Feb 2026 Seed2.0 Pro 94.2 AIME 2026
11 Feb 2026 GLM-5 92.7 AIME 2026 I
10 Feb 2026 Ring-2.5-1T 95 Best of AIME 26 (Mean@64) 93.75 / Heavy Thinking 95.00

20 system cards report it, from 13 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
19 Oct 2026 Muse Glimmer technical report 94.7 AIME 2026, high, 93.0 xhigh (Tables 2 and 5) Capabilities
10 Sep 2026 DeepSeek V4.1 Flash technical report 100 AIME 2026 (Pass@1) at reasoning effort 100 (Appendix B.3 text, Figure 12) Capabilities
5 Aug 2026 K-EXAONE 2.0 technical report 92.3 AIME 2026 (Table 6, Reasoning mode) Capabilities
31 Jul 2026 DiffusionGemma 26B A4B technical report 69.1 AIME 2026, TD with thinking, 50.8 no-think, 84.2 AR mode (Table 3) Capabilities
30 Jul 2026 Inkling-Small model card 95.5 AIME 2026 (5. Evaluations table) Capabilities
24 Jul 2026 Gemma 4 31B IT technical report 89.2 AIME 2026 no tools, thinking (Table 5) Capabilities
24 Jul 2026 Gemma 4 26B A4B technical report 88.3 AIME 2026 no tools, thinking (Table 5) Capabilities
22 Jul 2026 Solar Open 2 technical report 95.7 AIME2026, MathArena set (Table 3) Capabilities
15 Jul 2026 Inkling model card 97.1 AIME 2026 (5. Evaluations table, effort 0.99) Capabilities
13 Jun 2026 Ring-2.6-1T technical report 95.8 AIME 2026 (Avg@64), xhigh (Table 6). High gives 87.86 Capabilities
13 Jun 2026 Ling-2.6-1T technical report 87.4 AIME26, Mean@64, CoT (Table 5) Capabilities
9 Jun 2026 Keye-VL-2.0-30B-A3B technical report 93.3 AIME2026 (Figure 5, printed value) Capabilities
2 Jun 2026 MAI-Thinking-1 technical report 94.5 AIME 2026, pass@1, 256k max output tokens (Table 11) Capabilities
2 Jun 2026 MAI-Code-1-Flash model card 92.5 AIME 2026, Acc (Core reasoning table) Capabilities
26 May 2026 MiniMax-M2.7 technical report 94.2 AIME 2026, AA Index v4.0 protocol, pass@1 (Table 4) Capabilities
26 May 2026 MiniMax-M2.5 technical report 87.2 AIME 2026, AA Index v4.0 protocol, pass@1 (Table 4) Capabilities
9 Apr 2026 EXAONE 4.5 33B technical report 92.6 AIME 2026 (Table 3, Reasoning mode) Capabilities
19 Mar 2026 Nemotron-Cascade 2 30B-A3B technical report 90.9 AIME 2026, no tools (Table 1). TIR 95.0 Capabilities
17 Feb 2026 GLM-5 technical report 92.7 AIME 2026 I (Table 7) Capabilities
14 Feb 2026 Seed2.0 Pro model card 94.2 AIME 2026 (Table 3) 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 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 o3-mini
Grok 4 Heavy 100%
AIME 2026 Passed 85% in Feb 2026, by GPT-5.2
GPT-5.5 100%
Replaces AIME 2025

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%
  • 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%
  • 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%