Benchmark Almanac

HMMT Nov 2025

GPT-5.2 99.2% leads since Dec 2025 Fading

Models solve the 30 problems of the November 2025 Harvard-MIT Mathematics Tournament. Final answers are checked automatically against the answer key.

Made by MathArena (ETH Zurich SRI Lab) , problems by HMMT Released Nov 2025 Competition problems with answers All tasks public Last in a launch post Jun 2026 , dropped after it saturated Last model added Jun 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 matharena.ai
22
Models scored on its own leaderboard
19
Launch posts list it, from 9 labs
11
System cards report it, from 9 labs

GPT-5.2 leads with 99.2% 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:

22 models. Scroll sideways for the rest.

  1. 99.2 GPT-5.2
  2. 94.2 Step 3.5 Flash
  3. 94.2 GLM-5
  4. 93.3 Gemini 3 Pro
  5. 93.3 Grok 4.1 Fast
  6. 93.3 DeepSeek-V3.2-Speciale
  7. 93.3 Gemini 3 Flash
  8. 91.7 GLM-4.6
  9. 91.7 GPT-5.1
  10. 90.8 Grok 4 Fast
  11. 90.0 gpt-oss-120b
  12. 90.0 DeepSeek-V3.2
  13. 89.2 GPT-5
  14. 89.2 Kimi K2 Thinking
  15. 89.2 Kimi K2.5
  16. 88.3 Grok 4
  17. 84.2 GPT-5 mini
  18. 84.2 DeepSeek-V3.2-Exp
  19. 81.7 GPT-5 nano
  20. 80.0 Gemini 2.5 Pro
  21. 79.2 Falcon H1R 7B
  22. 75.0 QED-Nano
  1. 1 GPT-5.2 xhigh 99.2%
  2. 2 Step 3.5 Flash 94.2%
  3. 3 GLM-5 94.2%
  4. 4 Gemini 3 Pro 93.3%
  5. 5 Grok 4.1 Fast 93.3%
  6. 6 DeepSeek-V3.2-Speciale 93.3%
  7. 7 Gemini 3 Flash 93.3%
  8. 8 GLM-4.6 91.7%
  9. 9 GPT-5.1 high 91.7%
  10. 10 Grok 4 Fast 90.8%
  11. 11 gpt-oss-120b high 90%
  12. 12 DeepSeek-V3.2 thinking 90%
  13. 13 GPT-5 high 89.2%
  14. 14 Kimi K2 Thinking 89.2%
  15. 15 Kimi K2.5 thinking 89.2%
  16. 16 Grok 4 88.3%
  17. 17 GPT-5 mini high 84.2%
  18. 18 DeepSeek-V3.2-Exp thinking 84.2%
  19. 19 GPT-5 nano high 81.7%
  20. 20 Gemini 2.5 Pro Jun 2025 80%
  21. 21 Falcon H1R 7B 79.2%
  22. 22 QED-Nano 75%
#ModelScoreSettingDateSource
1 GPT-5.2 99.2% xhigh Dec 2025 Official leaderboard
2 Step 3.5 Flash 94.2% Feb 2026 Official leaderboard
3 GLM-5 94.2% Feb 2026 Official leaderboard
4 Gemini 3 Pro 93.3% Nov 2025 Official leaderboard
5 Grok 4.1 Fast 93.3% Nov 2025 Official leaderboard
6 DeepSeek-V3.2-Speciale 93.3% Dec 2025 Official leaderboard
7 Gemini 3 Flash 93.3% Dec 2025 Official leaderboard
8 GLM-4.6 91.7% Sep 2025 Official leaderboard
9 GPT-5.1 91.7% high Nov 2025 Official leaderboard
10 Grok 4 Fast 90.8% Sep 2025 Official leaderboard
11 gpt-oss-120b 90% high Aug 2025 Official leaderboard
12 DeepSeek-V3.2 90% thinking Dec 2025 Official leaderboard
13 GPT-5 89.2% high Aug 2025 Official leaderboard
14 Kimi K2 Thinking 89.2% Nov 2025 Official leaderboard
15 Kimi K2.5 89.2% thinking Jan 2026 Official leaderboard
16 Grok 4 88.3% Jul 2025 Official leaderboard
17 GPT-5 mini 84.2% high Aug 2025 Official leaderboard
18 DeepSeek-V3.2-Exp 84.2% thinking Sep 2025 Official leaderboard
19 GPT-5 nano 81.7% high Aug 2025 Official leaderboard
20 Gemini 2.5 Pro Jun 2025 80% Jun 2025 Official leaderboard
21 Falcon H1R 7B 79.2% Jan 2026 Official leaderboard
22 QED-Nano 75% Feb 2026 Official leaderboard

Labs report scores for 23 models in their cards. We plot every source in the charts below.

The best score, 99.2%, costs $0.19 a task

Runs on the cost frontier score higher than any run that's cheaper. Step 3.5 Flash comes within 5 points of the best for $0.01 a task, 14 times cheaper. The costs are the leaderboard's own estimates, so compare them only within this benchmark.

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 Cost per task, log scale GPT-5.2, 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-5.2, xhigh Step 3.5 Flash
ModelEffortHarnessScoreCost per taskOutput tokens
GPT-5.2 xhigh 99.2% $0.19 13,192
GPT-5.2 high 95.8% $0.14 10,015
Step 3.5 Flash 94.2% $0.01 45,001
GLM-5 94.2% $0.10 30,083
Gemini 3 Pro 93.3% $0.18 14,837
Grok 4.1 Fast 93.3% $0.005 10,401
DeepSeek-V3.2-Speciale 93.3% $0.01 25,249
Gemini 3 Flash 93.3% $0.06 20,413
GLM-4.6 91.7% $0.04 16,843
GPT-5.1 high 91.7% $0.20 19,593
Grok 4 Fast 90.8% $0.005 10,639
gpt-oss-120b high 90% $0.01 18,543
DeepSeek-V3.2 thinking 90% $0.007 17,165
GPT-5 high 89.2% $0.15 15,483
Kimi K2 Thinking 89.2% $0.07 28,752
Kimi K2.5 thinking 89.2% $0.08 26,085
Grok 4 88.3% $0.22 14,792
GPT-5 mini high 84.2% $0.03 14,859
DeepSeek-V3.2-Exp thinking 84.2% $0.007 17,348
GPT-5 nano high 81.7% $0.01 26,955
Gemini 2.5 Pro Jun 2025 80% $0.22 21,190
Falcon H1R 7B 79.2% $0.006 28,381
QED-Nano 75% 39,270

3 models have held number one since its release in Nov 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.

Z.ai Google DeepMind OpenAI DeepSeek Qwen Ant Group
Own leaderboard Passed 85% in Nov 2025, by GLM-4.6
GPT-5.2 99.2%
Lab claims Passed 85% in Dec 2025, by DeepSeek-V3.2-Speciale
Ring-2.5-1T 98.3%

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% 2025 2026 Near the ceiling (85%) GPT-5.2 DeepSeek-V3.2-Speciale gpt-oss-120b Gemini 2.5 Pro 0% 20% 40% 60% 80% 100% 2025 2026 Near the ceiling (85%) GPT-5.2 GLM-4.6

Labs headline it less often now, last in Jun 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
Z.ai
LG AI Research
Upstage
Qwen
StepFun
Ant Group
ByteDance Seed
Sarvam AI
DateLaunch postScoreAs the post writes it
16 Jun 2026 GLM-5.2 94.4 HMMT Nov. 2025
22 Apr 2026 Qwen3.6 27B 90.7 HMMT Nov 25
15 Apr 2026 Qwen 3.6 35B-A3B 89.1 HMMT Nov 25
7 Apr 2026 GLM-5.1 94 HMMT Nov. 2025
2 Apr 2026 Qwen 3.6 Plus 94.6 HMMT Nov 25
30 Mar 2026 Qwen3.5-Omni-Plus 84.4 HMMT Nov 25
6 Mar 2026 Sarvam 30B 74.2 HMMT Nov 2025
6 Mar 2026 Sarvam 105B 85.8 HMMT (Nov 25)
24 Feb 2026 Qwen3.5-122B-A10B 90.3 HMMT Nov 25
16 Feb 2026 Qwen3.5 397B-A17B 92.7 HMMT Nov 25
14 Feb 2026 Seed2.0 Pro 93.3 HMMT Nov 2025
11 Feb 2026 GLM-5 96.9 HMMT Nov. 2025
10 Feb 2026 Ring-2.5-1T 98.3 Best of HMMT (Nov. 25) (Mean@64) 97.34 / Heavy Thinking 98.33
2 Feb 2026 Step 3.5 Flash 94 HMMT 2025 (Nov.)
23 Jan 2026 Qwen3-Max 94.7 HMMT Nov 25
31 Dec 2025 Solar Open 100B 80 HMMT 2025 (Nov)
31 Dec 2025 K-EXAONE 86.8 HMMT Nov 2025
22 Dec 2025 GLM-4.7 93.5 HMMT Nov. 2025
1 Dec 2025 DeepSeek-V3.2 90.2 HMMT Nov 2025 (Pass@1)

11 system cards report it, from 9 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
13 Jun 2026 Ling-2.6-1T technical report 81.9 HMMT-Nov25, Mean@64, CoT (Table 5) Capabilities
17 Apr 2026 Qwen3.5-Omni-Plus technical report 84.4 HMMT Nov 25 (Table 4) Capabilities
28 Feb 2026 qwen3-coder-next technical report 75.6 HMMT25 Nov (Table 9) Capabilities
17 Feb 2026 GLM-5 technical report 96.9 HMMT Nov. 2025 (Table 7) Capabilities
14 Feb 2026 Seed2.0 Pro model card 93.3 HMMT Nov 2025 (Table 3) Capabilities
11 Feb 2026 Step 3.5 Flash technical report 94 HMMT 2025 Nov., avg of 64 (Table 5). PaCoRe 97.8, with Python 98.0 (Table 12) Capabilities
30 Jan 2026 Kimi K2.5 technical report 91.1 HMMT Nov 2025, average 24k output tokens (Table 5) Capabilities
11 Jan 2026 Solar Open 100B technical report 80 HMMT 2025 (Nov) (Table 6) Capabilities
5 Jan 2026 K-EXAONE technical report 86.8 HMMT Nov 2025 (Table 3, Reasoning mode). Non-reasoning 43.2 (Table 4) Capabilities
1 Dec 2025 DeepSeek-V3.2 technical report 90.2 HMMT Nov 2025 (Pass@1), DeepSeek-V3.2 Thinking (Table 2) Capabilities
1 Dec 2025 DeepSeek-V3.2-Speciale technical report 94.4 HMMT Nov 2025 (Pass@1), DeepSeek-V3.2-Speciale (Table 3) Capabilities

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%