Benchmark Almanac

HMMT 2026

GPT-5.5 98.5% leads since Apr 2026 In use

Models solve the 33 final-answer problems from the February 2026 Harvard-MIT Mathematics Tournament. Scored by exact answer match.

An exam, not a dedicated benchmark. Top models score above 95%.

Made by HMMT (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
16
Launch posts list it, from 10 labs
7
System cards report it, from 6 labs

GPT-5.5 leads with 98.5% 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. 98.5 GPT-5.5
  2. 97.7 GPT-5.4
  3. 97.0 GPT-5.2
  4. 97.0 Kimi K3
  5. 96.2 Claude Opus 4.6
  6. 95.5 Gemini 3.5 Flash
  7. 95.5 Claude Opus 4.8
  8. 94.7 Gemini 3.1 Pro
  9. 94.7 Kimi K2.6
  10. 93.9 Claude Opus 4.7
  11. 93.9 DeepSeek-V4-Flash
  12. 93.9 DeepSeek-V4-Pro
  13. 92.4 GLM-5.2
  14. 89.4 Gemini 3 Flash
  15. 89.4 GLM-5.1
  16. 89.4 Gemini 3.6 Flash
  17. 87.9 Qwen3.5 397B-A17B
  18. 87.9 Step 3.7 Flash
  19. 87.1 Kimi K2.5
  20. 86.4 Gemini 3 Pro
  21. 86.4 Grok 4.1 Fast
  22. 86.4 Step 3.5 Flash
  23. 86.4 GLM-5
  24. 84.8 Nemotron 3 Super
  25. 84.1 DeepSeek-V3.2
  26. 81.8 Qwen3.5-35B-A3B
  27. 81.1 Qwen3.5-27B
  28. 78.8 Qwen3-30B-A3B-Thinking
  29. 71.2 Qwen3.5-9B
  30. 64.4 QED-Nano
  31. 53.0 Qwen3-4B-Thinking-2507
  1. 1 GPT-5.5 xhigh 98.5%
  2. 2 GPT-5.4 xhigh 97.7%
  3. 3 GPT-5.2 high 97%
  4. 4 Kimi K3 thinking 97%
  5. 5 Claude Opus 4.6 high 96.2%
  6. 6 Gemini 3.5 Flash 95.5%
  7. 7 Claude Opus 4.8 max 95.5%
  8. 8 Gemini 3.1 Pro 94.7%
  9. 9 Kimi K2.6 thinking 94.7%
  10. 10 Claude Opus 4.7 xhigh 93.9%
  11. 11 DeepSeek-V4-Flash max 93.9%
  12. 12 DeepSeek-V4-Pro max 93.9%
  13. 13 GLM-5.2 92.4%
  14. 14 Gemini 3 Flash 89.4%
  15. 15 GLM-5.1 89.4%
  16. 16 Gemini 3.6 Flash 89.4%
  17. 17 Qwen3.5 397B-A17B 87.9%
  18. 18 Step 3.7 Flash 87.9%
  19. 19 Kimi K2.5 thinking 87.1%
  20. 20 Gemini 3 Pro 86.4%
  21. 21 Grok 4.1 Fast 86.4%
  22. 22 Step 3.5 Flash 86.4%
  23. 23 GLM-5 86.4%
  24. 24 Nemotron 3 Super 84.8%
  25. 25 DeepSeek-V3.2 thinking 84.1%
  26. 26 Qwen3.5-35B-A3B 81.8%
  27. 27 Qwen3.5-27B 81.1%
  28. 28 Qwen3-30B-A3B-Thinking Jul 2025 78.8%
  29. 29 Qwen3.5-9B 71.2%
  30. 30 QED-Nano 64.4%
  31. 31 Qwen3-4B-Thinking-2507 53%
#ModelScoreSettingDateSource
1 GPT-5.5 98.5% xhigh Apr 2026 Official leaderboard
2 GPT-5.4 97.7% xhigh Mar 2026 Official leaderboard
3 GPT-5.2 97% high Dec 2025 Official leaderboard
4 Kimi K3 97% thinking Jul 2026 Official leaderboard
5 Claude Opus 4.6 96.2% high Feb 2026 Official leaderboard
6 Gemini 3.5 Flash 95.5% May 2026 Official leaderboard
7 Claude Opus 4.8 95.5% max May 2026 Official leaderboard
8 Gemini 3.1 Pro 94.7% Feb 2026 Official leaderboard
9 Kimi K2.6 94.7% thinking Apr 2026 Official leaderboard
10 Claude Opus 4.7 93.9% xhigh Apr 2026 Official leaderboard
11 DeepSeek-V4-Flash 93.9% max Apr 2026 Official leaderboard
12 DeepSeek-V4-Pro 93.9% max Apr 2026 Official leaderboard
13 GLM-5.2 92.4% Jun 2026 Official leaderboard
14 Gemini 3 Flash 89.4% Dec 2025 Official leaderboard
15 GLM-5.1 89.4% Apr 2026 Official leaderboard
16 Gemini 3.6 Flash 89.4% Jul 2026 Official leaderboard
17 Qwen3.5 397B-A17B 87.9% Feb 2026 Official leaderboard
18 Step 3.7 Flash 87.9% May 2026 Official leaderboard
19 Kimi K2.5 87.1% thinking Jan 2026 Official leaderboard
20 Gemini 3 Pro 86.4% Nov 2025 Official leaderboard
21 Grok 4.1 Fast 86.4% Nov 2025 Official leaderboard
22 Step 3.5 Flash 86.4% Feb 2026 Official leaderboard
23 GLM-5 86.4% Feb 2026 Official leaderboard
24 Nemotron 3 Super 84.8% Mar 2026 Official leaderboard
25 DeepSeek-V3.2 84.1% thinking Dec 2025 Official leaderboard
26 Qwen3.5-35B-A3B 81.8% Feb 2026 Official leaderboard
27 Qwen3.5-27B 81.1% Feb 2026 Official leaderboard
28 Qwen3-30B-A3B-Thinking Jul 2025 78.8% Jul 2025 Official leaderboard
29 Qwen3.5-9B 71.2% Feb 2026 Official leaderboard
30 QED-Nano 64.4% Feb 2026 Official leaderboard
31 Qwen3-4B-Thinking-2507 53% Aug 2025 Official leaderboard

Labs report scores for 18 models in their cards. We plot every source in the charts below.

The best score, 98.5%, costs $0.26 a task

Runs on the cost frontier score higher than any run that's cheaper. DeepSeek-V4-Flash at max comes within 4.5 points of the best for $0.01 a task, 17 times cheaper. The costs are the leaderboard's own estimates, so compare them only within this benchmark.

xAI OpenAI Qwen Moonshot AI DeepSeek Google DeepMind 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 Kimi K2.6, thinking DeepSeek-V4-Flash, max Grok 4.1 Fast Qwen3.5-9B 0% 20% 40% 60% 80% 100% $0.001 $0.01 $0.10 $1.00 Cost per task, log scale GPT-5.5, xhigh Grok 4.1 Fast
ModelEffortHarnessScoreCost per taskOutput tokens
GPT-5.5 xhigh 98.5% $0.26 8,496
GPT-5.4 xhigh 97.7% $0.22 14,538
GPT-5.2 high 97% $0.19 13,709
Kimi K3 thinking 97% $0.22 14,694
Claude Opus 4.6 high 96.2% $0.64 25,758
Gemini 3.5 Flash 95.5% $0.15 16,121
Claude Opus 4.8 max 95.5% $0.76 30,441
Gemini 3.1 Pro 94.7% $0.20 16,761
Kimi K2.6 thinking 94.7% $0.13 33,563
Claude Opus 4.7 xhigh 93.9% $0.56 22,279
DeepSeek-V4-Flash max 93.9% $0.01 54,365
DeepSeek-V4-Pro max 93.9% $0.14 40,696
GLM-5.2 92.4% $0.10 23,293
Gemini 3 Flash 89.4% $0.07 21,896
GLM-5.1 89.4% $0.13 39,645
Gemini 3.6 Flash 89.4% $0.10 13,542
Qwen3.5 397B-A17B 87.9% $0.09 24,189
Step 3.7 Flash 87.9% $0.02 81,069
Kimi K2.5 thinking 87.1% $0.09 28,765
Gemini 3 Pro 86.4% $0.19 15,502
Grok 4.1 Fast 86.4% $0.007 13,846
Step 3.5 Flash 86.4% $0.02 60,004
GLM-5 86.4% $0.11 33,206
Nemotron 3 Super 84.8% 45,773
DeepSeek-V3.2 thinking 84.1% $0.01 24,288
Qwen3.5-35B-A3B 81.8% $0.06 42,716
Qwen3.5-27B 81.1% $0.06 37,594
Qwen3-30B-A3B-Thinking Jul 2025 78.8% $0.007 24,319
Qwen3.5-9B 71.2% $0.006 40,607
QED-Nano 64.4% 57,709
Qwen3-4B-Thinking-2507 53% $0.008 27,600

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 Upstage Qwen Moonshot AI DeepSeek
Own leaderboard Passed 85% in Feb 2026, by GPT-5.2
GPT-5.5 98.5%
Lab claims Passed 85% in Apr 2026, by Qwen 3.6 Plus
Qwen3.7-Max 97.1%

The best score passed 85% in Nov 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 Gemini 3 Pro Qwen3-30B-A3B-Thinking 0% 20% 40% 60% 80% 100% 2026 2027 Near the ceiling (85%) GPT-5.5 Gemini 3 Pro

8 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.

Upstage
Qwen
Z.ai
Moonshot AI
DeepSeek
Microsoft
Thinking Machines
LG AI Research
Shanghai AI Lab
iFlytek
DateLaunch postScoreAs the post writes it
22 Oct 2026 Spark-X2.6-4B 81.3 HMMT Feb 26
13 Sep 2026 Intern-S2-397B 93.6 HMMT-2026
31 Jul 2026 K-EXAONE 2.0 78.4 HMMT Feb 2026
30 Jul 2026 Inkling-Small 90.2 HMMT Feb 2026
22 Jul 2026 Solar Open 2 93.9 HMMT2602 (MathArena)
16 Jun 2026 GLM-5.2 92.5 HMMT Feb. 2026
2 Jun 2026 MAI-Thinking-1 84.9 HMMT Feb 2026
1 Jun 2026 Qwen3.7-Plus 92.9 HMMT 2026 Feb
20 May 2026 Qwen3.7-Max 97.1 HMMT 2026 Feb
24 Apr 2026 DeepSeek-V4-Pro 95.2 HMMT 2026 Feb (Pass@1)
22 Apr 2026 Qwen3.6 27B 84.3 HMMT Feb 26
20 Apr 2026 Kimi K2.6 92.7 HMMT 2026 (Feb)
15 Apr 2026 Qwen 3.6 35B-A3B 83.6 HMMT Feb 26
7 Apr 2026 GLM-5.1 82.6 HMMT Feb. 2026
2 Apr 2026 Qwen 3.6 Plus 87.8 HMMT Feb 26
24 Mar 2026 Solar Pro 3 (March 2026 update) 65.9 HMMT'26

7 system cards report it, from 6 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
5 Aug 2026 K-EXAONE 2.0 technical report 78.4 HMMT Feb 2026 (Table 6, Reasoning mode) Capabilities
30 Jul 2026 Inkling-Small model card 90.2 HMMT Feb 2026 (5. Evaluations table) Capabilities
22 Jul 2026 Solar Open 2 technical report 93.9 HMMT2602, MathArena set (Table 3) Capabilities
13 Jun 2026 Ring-2.6-1T technical report 93.5 HMMT-Feb26 (Avg@64), xhigh (Table 6). High gives 67.80 Capabilities
2 Jun 2026 MAI-Thinking-1 technical report 84.9 HMMT Feb 2026, pass@1, 256k max output tokens (Table 11) Capabilities
26 Apr 2026 DeepSeek-V4-Pro technical report 95.2 HMMT 2026 Feb (Pass@1), DS-V4-Pro Max (Table 6) Capabilities
26 Apr 2026 DeepSeek-V4-Flash technical report 94.8 HMMT 2026 Feb (Pass@1), DeepSeek-V4-Flash Max (Table 7) 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 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%
  • 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%