Benchmark Almanac

USAMO 2026

GPT-5.5 98.2% leads since April 2026
GPT-5.5 98.2% leads since April 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.

MathArena reports GPT-5.4 at 95.2% and says it essentially saturates the benchmark.

Released
March 2026
Kind
Written proofs
Level
Competition
Scored
Graded by people
Answers
Written proof
Tasks
All public
Strongest model tested
GPT-5.5 #20 on the capability index
Last in a launch post
July 2026
Last model added
July 2026
Scores refreshed
November 12th, 2026
9
Models scored on MathArena’s leaderboard
3
Launch posts list it, from 3 labs
5
System cards report it, from 2 labs

GPT-5.5 leads with 98.2% among models released by April 2026

Best score per model from MathArena’s 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 98.2%, 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 April 2026, and nobody has tested a newer one. Among the 20 highest-ranked models on the capability index, these have no score yet:

  1. 98.2 GPT-5.5
  2. 95.2 GPT-5.4
  3. 74.4 Gemini 3.1 Pro
  4. 60.7 DeepSeek-V4-Pro
  5. 51.2 Kimi K2.6
  6. 47.0 Claude Opus 4.6
  7. 44.6 Step 3.5 Flash
  8. 36.3 Qwen3.5 397B-A17B
  9. 35.1 GLM-5
  1. 1 GPT-5.5 xhigh 98.2%
  2. 2 GPT-5.4 xhigh 95.2%
  3. 3 Gemini 3.1 Pro high 74.4%
  4. 4 DeepSeek-V4-Pro max 60.7%
  5. 5 Kimi K2.6 high 51.2%
  6. 6 Claude Opus 4.6 high 47%
  7. 7 Step 3.5 Flash 44.6%
  8. 8 Qwen3.5 397B-A17B 36.3%
  9. 9 GLM-5 35.1%

Source: MathArena’s leaderboard

Sources: MathArena’s leaderboard and labs’ model cards

#ModelScoreSettingDateSource
1 GPT-5.5 98.2% xhigh Apr 2026 Official leaderboard
2 GPT-5.4 95.2% xhigh Mar 2026 Official leaderboard
3 Gemini 3.1 Pro 74.4% high Feb 2026 Official leaderboard
4 DeepSeek-V4-Pro 60.7% max Apr 2026 Official leaderboard
5 Kimi K2.6 51.2% high Apr 2026 Official leaderboard
6 Claude Opus 4.6 47% high Feb 2026 Official leaderboard
7 Step 3.5 Flash 44.6% Feb 2026 Official leaderboard
8 Qwen3.5 397B-A17B 36.3% Feb 2026 Official leaderboard
9 GLM-5 35.1% Feb 2026 Official leaderboard
#ModelScoreSettingDateSource
1 GPT-5.5 98.2% xhigh Apr 2026 Official leaderboard
2 Claude Mythos Preview 97.6% Apr 2026 Lab model card
3 Nemotron 3 Ultra 97.6% Jun 2026 Lab model card
4 Claude Opus 4.8 96.7% May 2026 Lab model card
5 GPT-5.4 95.2% xhigh Mar 2026 Official leaderboard
6 Claude Sonnet 5 79.5% Jun 2026 Lab model card
7 Gemini 3.1 Pro 74.4% high Feb 2026 Official leaderboard
8 Tencent Hy3 72% Jul 2026 Lab model card
9 Claude Opus 4.7 69.3% Apr 2026 Lab model card
10 DeepSeek-V4-Pro 60.7% max Apr 2026 Official leaderboard
11 Kimi K2.6 51.2% high Apr 2026 Official leaderboard
12 Claude Opus 4.6 47% high Feb 2026 Official leaderboard
13 Step 3.5 Flash 44.6% Feb 2026 Official leaderboard
14 Qwen3.5 397B-A17B 36.3% Feb 2026 Official leaderboard
15 GLM-5 35.1% Feb 2026 Official leaderboard

Labs report scores for 6 models in their cards. Switch the scores above to rank them together.

The best score, 98.2%, costs $0.79 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.

StepFun Google DeepMind Moonshot AI OpenAI Cost frontier One model at each effort level
0% 20% 40% 60% 80% 100% $0.01 $0.10 $1.00 $10 Cost per task, log scale GPT-5.5, xhigh Gemini 3.1 Pro, high Kimi K2.6, high Step 3.5 Flash 0% 20% 40% 60% 80% 100% $0.01 $0.10 $1.00 $10 Cost per task, log scale GPT-5.5, xhigh Gemini 3.1 Pro, high

Source: MathArena’s leaderboard

ModelEffortHarnessScoreCost per taskOutput tokens
GPT-5.5 xhigh 98.2% $0.79 26,399
GPT-5.4 xhigh 95.2% $0.86 56,878
Gemini 3.1 Pro high 74.4% $0.37 30,598
DeepSeek-V4-Pro max 60.7% $0.50 143,526
Kimi K2.6 high 51.2% $0.25 63,178
Claude Opus 4.6 high 47% $2.21 87,828
Step 3.5 Flash 44.6% $0.04 124,206
Qwen3.5 397B-A17B 36.3% $0.12 33,356
GLM-5 35.1% $0.24 76,404

3 models have held number one on MathArena’s leaderboard since its release in March 2026 4 models have held number one with claims included since its release in March 2026

Lab claims count from the day the card came out, independent runs from the model’s release. We draw one row each for MathArena’s leaderboard and what labs claim in their cards.

Google DeepMind OpenAI Anthropic
MathArena leaderboard Passed 85% in Mar 2026, by GPT-5.4
GPT-5.5 98.2%
Lab claims Passed 85% in Apr 2026, by Claude Mythos Preview
Claude Mythos Preview 97.6%

Sources: MathArena’s leaderboard and labs’ model cards

GPT-5.4 on the official leaderboard passed 85% in March 2026

All the scores we have, one dot each, so a model tested at several settings appears more than once.

On MathArena’s 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 GPT-5.4 Gemini 3.1 Pro Claude Opus 4.6 0% 20% 40% 60% 80% 100% 2026 Near the ceiling (85%) GPT-5.5 Gemini 3.1 Pro

Sources: MathArena’s leaderboard and labs’ model cards

2 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.

Anthropic
MiniMax
Tencent
DateLaunch postScoreAs the post writes it
Jul 2, 2026 Tencent Hy3 72 USAMO 2026
Jun 1, 2026 MiniMax-M3 36 USAMO 2026
Apr 7, 2026 Claude Mythos Preview 97.6 USAMO 2026

5 system cards report it, from 2 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
Jun 30, 2026 Claude Sonnet 5 system card 79.5 USAMO 2026, high effort, 300k token limit, average over 10 attempts per problem, MathArena-style grading (8.6) Capabilities
Jun 9, 2026 Nemotron 3 Ultra technical report 97.6 USAMO 2026, 41/42, generate-verify-refine test-time scaling, human graded (Table 11) Capabilities
May 28, 2026 Claude Opus 4.8 system card 96.7 USAMO 2026, MathArena grading with the minimum of three judges, high effort, 300k token limit, 10 attempts per problem (8.7) Capabilities
Apr 16, 2026 Claude Opus 4.7 system card 69.3 USAMO 2026, MathArena grading, medium effort, 300k token limit, average of 10 attempts per problem (8.6) Capabilities
Apr 7, 2026 Claude Mythos Preview system card 97.6 USAMO 2026, max effort, no tools, 10 trials per problem, MathArena grading with the minimum of 3 judges (Table 6.3.A, 6.8) 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.

DeepSeek Google DeepMind OpenAI
USAMO 2026 Passed 85% in Mar 2026, by GPT-5.4
GPT-5.5 98.2%
USAMO 2025 Not tested since May 2025
DeepSeek-R1 30.1%
USAMO 2026 Passed 85% in Mar 2026, by GPT-5.4
GPT-5.5 98.2%

Source: Official leaderboards

Replaces USAMO 2025

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%
  • MathArena Apex Shortlist 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. GPT-5.5 98.4%