Why this benchmark is useful
Historical contest-math signal — once useful for hard integer reasoning. Solved for frontier models with extended thinking; keep for context, not current ranking.
American Invitational Mathematics Examination
AIME
Hard high-school competition math: 15 integer-answer problems per exam. Adopted as a frontier reasoning eval because each year's fresh paper resists contamination — until it leaks.
What this does not measure
- Tiny sample (15 problems/year) — one problem is ~6.7%, so single-run scores are noisy; demand pass@k or multiple seeds.
- General capability — strong AIME scores correlate with reasoning training, not with everyday helpfulness.
- Which year — '90% on AIME' is meaningless without the exam year and number of attempts.
Coverage map
- Task family
- Math
- Format
- 15 problems, integer answers 0-999; reported as accuracy on AIME 2024/2025 papers.
- Scoring
- Accuracy (often pass@1 or cons@64).
- Maintainer
- MAA (exam) · used as an LLM eval
How to read the scores
Do not use near-100% AIME headlines to separate today's frontier. Tiny yearly samples (15 problems) are noisy; always note exam year and attempt count.
What it does not cover
- Tiny sample (15 problems/year) — one problem is ~6.7%, so single-run scores are noisy; demand pass@k or multiple seeds.
- General capability — strong AIME scores correlate with reasoning training, not with everyday helpfulness.
- Which year — '90% on AIME' is meaningless without the exam year and number of attempts.
Scores
37 rowsEvidence ledger
3737 rows
98.67%best score
31source-checked
3sources
2026-09-01to 2024-09-12
Where the rows land
0255075100
Normalized to this benchmark's axis (0–100). Open the table for raw units.
Newest receipts
- GPT-5 Codex (high)98.67% ·
- Gemini 3 Flash Preview (Reasoning)97% ·
- GPT-5.2 (medium)96.67% ·
- MiMo-V2-Flash (Reasoning)96.33% ·
- Gemini 3 Pro95.67% ·
Scores use this benchmark's own unit and axis, not a universal quality score.
| # | Model | Release date | Score | Provenance | Trust |
|---|---|---|---|---|---|
| 1 | GPT-5 Codex (high)OpenAI | 2025-09-23 | 98.67% | 2026-09-01 | Source-checked |
| 2 | Gemini 3 Flash Preview (Reasoning)Google | 2025-12-17 | 97% | 2026-09-01 | Source-checked |
| 3 | GPT-5.2 (medium)OpenAI | 2025-12-11 | 96.67% | 2026-09-01 | Source-checked |
| 4 | MiMo-V2-Flash (Reasoning)Xiaomi | 2025-12-16 | 96.33% | 2026-09-01 | Source-checked |
| 5 | Gemini 3 ProGoogle | 2025-11-18 | 95.67% | 2026-09-01 | Source-checked |
| 6 | GPT-5.1 Codex (high)OpenAI | 2025-11-13 | 95.67% | 2026-09-01 | Source-checked |
| 7 | GLM-4.7 (Reasoning)Zhipu | 2025-12-22 | 95% | 2026-09-01 | Source-checked |
| 8 | Kimi K2 ThinkingMoonshot | 2025-11-06 | 94.67% | 2026-09-01 | Source-checked |
| 9 | GPT-5 (high)OpenAI | 2025-08-07 | 94.33% | 2026-09-01 | Source-checked |
| 1 | Claude Opus 4.8AnthropicAIME 2024 with extended thinking. | 2026-05-28 | 94.2% | Anthropic Claude 4 family2026-05-28 | Needs audit |
| 11 | GPT-5.1OpenAI | 2025-11-13 | 94% | 2026-09-01 | Source-checked |
| 2 | Claude 3.7 Sonnet (2025-02)AnthropicAIME 2024 with extended thinking enabled. | 2025-02-19 | 93.7% | Anthropic Claude 3.7 Sonnet announcement2025-02-19 | Source-checked |
| 13 | Grok 4xAI | 2025-07-10 | 92.67% | 2026-09-01 | Source-checked |
| 14 | DeepSeek V3.2 (Reasoning)DeepSeek | 2025-12-01 | 92% | 2026-09-01 | Source-checked |
| 3 | GPT 5.5OpenAI | 2026-04-23 | 91.8% | OpenAI model release notes2026-04-23 | Needs audit |
| 16 | GPT-5 (medium)OpenAI | 2025-08-07 | 91.67% | 2026-09-01 | Source-checked |
| 17 | Claude Opus 4.5 (Reasoning)Anthropic | 2025-11-24 | 91.33% | 2026-09-01 | Source-checked |
| 4 | GPT 5OpenAI | 2025-12-11 | 89.2% | OpenAI model release notes2025-12-01 | Needs audit |
| 5 | DeepSeek R1DeepSeekAIME 2024 with chain-of-thought. | 2025-01-20 | 88.4% | DeepSeek R1 model card2025-01-20 | Needs audit |
| 20 | OpenAI o3OpenAI | 2025-04-16 | 88.33% | 2026-09-01 | Source-checked |
| 21 | Claude 4.5 Sonnet (Reasoning)Anthropic | 2025-09-29 | 88% | 2026-09-01 | Source-checked |
| 6 | Gemini 3.1 ProGoogle | 2026-02-19 | 87.6% | Google Gemini 3.1 Pro2026-02-19 | Needs audit |
| 7 | o3 (cons@64, AIME 2025)OpenAIAIME 2025 paper; 64-sample consensus. Single-shot is markedly lower. | 2025-04-16 | 87.5% | OpenAI: Learning to Reason2025-01-20 | Source-checked |
| 24 | Gemini 3 Pro Preview (low)Google | 2025-11-18 | 86.67% | 2026-09-01 | Source-checked |
| 9 | Kimi K2.6Moonshot | 2026-04-20 | 85.2% | Moonshot Kimi K2.62026-04-20 | Needs audit |
| 10 | OpenAI o1 (AIME 2024, cons@64)OpenAIWith 64-sample consensus; single-shot is markedly lower. | 2024-09-12 | 83.3% | OpenAI: Learning to Reason2024-09-12 | Source-checked |
| 27 | GPT-5 (low)OpenAI | 2025-08-07 | 83% | 2026-09-01 | Source-checked |
| 28 | MiniMax-M2.1MiniMax | 2025-12-23 | 82.67% | 2026-09-01 | Source-checked |
| 29 | Claude 4.1 Opus (Reasoning)Anthropic | 2025-08-05 | 80.33% | 2026-09-01 | Source-checked |
| 30 | Claude 4 Opus (Reasoning)Anthropic | 2025-05-22 | 73.33% | 2026-09-01 | Source-checked |
| 31 | Claude Opus 4.5Anthropic | 2025-11-24 | 62.67% | 2026-09-01 | Source-checked |
| 32 | Gemini 3 Flash PreviewGoogle | 2025-12-17 | 55.67% | 2026-09-01 | Source-checked |
| 33 | Mistral Large 3Mistral | 2025-12-02 | 38% | 2026-09-01 | Source-checked |
| 34 | Llama 4 MaverickMeta | 2026-04-05 | 19.33% | 2026-09-01 | Source-checked |
| 35 | Llama 4 ScoutMeta | 2026-04-05 | 14% | 2026-09-01 | Source-checked |
| 36 | Command ACohere | 2026-03-15 | 13% | 2026-09-01 | Source-checked |
| 37 | Claude 3 OpusAnthropic | 2024-03-04 | 3.33% | 2026-09-01 | Source-checked |
What it covers
Archived — GPT-5.2 and peers now report 100% on recent AIME papers with extended thinking. Treat as solved, not a frontier separator. Prefer /benchmarks/frontiermath.
Where it breaks down
No ceiling note is recorded yet. Treat clustering near the top as a warning that the benchmark may no longer separate frontier models.
Receipts
Sources and further reading
- OpenAI o1 announcement· checked 2026-05-30
Questions about this benchmark
Every answer below is assembled from the dated fields on this page. Nothing is written separately for search.
What does AIME measure?
Hard high-school competition math: 15 integer-answer problems per exam. Adopted as a frontier reasoning eval because each year's fresh paper resists contamination — until it leaks.
What does a high AIME score not prove?
A strong AIME result says nothing about:
- Tiny sample (15 problems/year) — one problem is ~6.7%, so single-run scores are noisy; demand pass@k or multiple seeds.
- General capability — strong AIME scores correlate with reasoning training, not with everyday helpfulness.
- Which year — '90% on AIME' is meaningless without the exam year and number of attempts.
How is AIME scored?
Accuracy (often pass@1 or cons@64). Task format: 15 problems, integer answers 0-999; reported as accuracy on AIME 2024/2025 papers.
Is AIME saturated?
AIME is currently marked Defunct in the atlas.
Can AIME results be contaminated by training data?
Contamination risk for AIME is graded Medium contamination. Treat every row on this page as a public claim with a source and a date, not as a controlled experiment.
The Pack · Editorial newsletter
New tools in your inbox. Free.
One short email when a tool ships or changes status. No tracking, no third-party analytics. Unsubscribe in one click.