Metric
accuracy (%)
Status
Saturated
Category
Mathematics
Maintainer
MAA; curated by Epoch AI
Chance baseline
0%
What it measures
The American Invitational Mathematics Examination (AIME) measures advanced pre-college mathematical problem solving: multi-step, creative reasoning across algebra, geometry, number theory, and combinatorics. It is a human competition first, organized by the Mathematical Association of America (MAA) for students who have already excelled on the AMC 10 or AMC 12. As a language-model benchmark it has become one of the standard tests of mathematical reasoning, because its problems demand a genuine multi-step derivation rather than fact recall, and because each problem has a single integer answer that is trivial to grade automatically.
The task in detail
Each AIME is a 15-question, 3-hour test. Every answer is an integer between 000 and 999 inclusive, and no answer choices are provided, so a solver must actually produce the number. Calculators, rulers, protractors, and other aids are not allowed. Problems cover elementary algebra, geometry, trigonometry, number theory, probability, and combinatorics, and they escalate in difficulty through the paper, from approachable early questions to hard closing ones that stump most qualifiers. Between 2000 and 2026 the MAA offered two forms each year, AIME I and AIME II (the second an alternate for students who could not sit the first), of equal standing. The AIME sits between the AMC and the USA Mathematical Olympiad (USAMO or USAJMO) in the qualification chain.
For language models the exam is used almost verbatim: the model is given a problem statement and must output the integer answer, usually after chain-of-thought working. Because a full AIME is only 15 problems, and only about 30 fresh problems appear per year, evaluators either aggregate several years, run each problem many times, or use held-out "mock" sets. Epoch AI, for example, hosts a Mock AIME 2024-2025 assembled from 45 problems across three OTIS mock exams (15 from Mock AIME 2024, 15 from Mock AIME 2025 I, and 15 from Mock AIME 2025 II), positioned as harder than the older MATH Level 5 benchmark but easier than research-level sets.
How models are scored
The dominant metric is exact-match accuracy, often reported as pass@1: the fraction of problems whose final integer answer exactly matches the key, after the answer is extracted from the model's output. There is no partial credit, mirroring the human contest, where one point is earned per correct answer, no points are deducted for wrong ones, and scores are integers from 0 to 15.
Because a single run of a probabilistic model is noisy on so few problems, careful evaluators sample each problem many times and average. Epoch AI runs most models 16 times on its Mock AIME set, uses each model's API-default temperature, zero-shot chain-of-thought prompting, and an empty system prompt, and reports a mean with a confidence interval of plus or minus one standard error. Other groups report consensus (majority-vote) accuracy or best-of-n, which raise the number but measure something different from a single attempt, so the sampling method always has to be read alongside the score.
How to read a score
Always check three things: which year's problems, how many samples, and whether the number is pass@1 or a consensus or majority-vote figure. A "90% on AIME" from majority voting over dozens of samples is not comparable to 90% from a single attempt. Because there are only 15 problems per exam, one problem is worth nearly 7 percentage points, so single-run scores on one year are statistically shaky and confidence intervals matter.
Contamination is the other lens. If a model may have seen a given year's problems and solutions during training, a high score can reflect memorization rather than reasoning, which is why fresh exams and mock sets are used as a sanity check. Treat a high AIME number as strong evidence of reasoning mainly when it holds up on problems released after the model's training cutoff.
History and notable results
The AIME has been a fixture of American math competitions for decades, but it became a marquee AI benchmark in 2024. For most of the language-model era it was a hard target: general chat models solved only a small fraction of problems. That changed sharply with the arrival of dedicated reasoning models. OpenAI presented its first o1 model in late 2024 with a large reported jump on AIME 2024 over the previous GPT-4o generation, and the result was widely cited as a turning point for test-time reasoning.
Through 2025, frontier reasoning systems pushed AIME accuracy to very high levels, to the point that Epoch AI notes older math benchmarks are approaching saturation and added a harder Mock AIME set to keep resolving differences between top models. Because the standings move with every release, the live leaderboard is the place to see current results. The exam itself is not static either: the two-form structure of AIME I and AIME II ran from 2000 through 2026, after which the MAA is changing the exam's format.
Limitations and criticism
The biggest issue is data contamination: past AIME problems and their solutions are all over the web, so a model may have effectively memorized them. The tiny size (15 problems per form) makes single-run scores noisy and easy to over-interpret. The integer-answer format means a model can occasionally reach the right number through a flawed or lucky route, which exact-match scoring cannot detect. And scores are sensitive to prompt and sampling choices, so headline numbers from different labs are often not directly comparable.
Related benchmarks
Epoch AI's Mock AIME 2024-2025 is the closest sibling, a harder, less-contaminated variant built from mock exams. The MATH dataset and its hardest MATH Level 5 tier draw on AIME and AMC problems for a broader test of the same skills. Above AIME sit olympiad-level challenges built from USAMO and IMO problems, along with research-grade sets such as FrontierMath, while general reasoning suites like GPQA Diamond cover scientific rather than purely mathematical reasoning.
References
Leaderboard
Checked Nov 11, 2026, 07:00 UTC · scores last changed Nov 11, 2026Best AIME score per model, ranked. Each score links to its source.
| # | Model | Score | Source |
|---|---|---|---|
| 1 | Google: Gemini 4 Argon(high)Google DeepMind | 100% | epoch |
| 2 | gpt-5.5-pre-release_xhigh(xhigh)OpenAI | 100% | epoch |
| 3 | gpt-5.5-pro-pre-release_xhigh(xhigh)OpenAI | 100% | epoch |
| 4 | Qwen: Qwen3.8 Max (0902)(xhigh)Alibaba | 100% | epoch |
| 5 | Anthropic: Claude Fable 5.1(max)Anthropic | 100% | epoch |
| 6 | OpenAI: GPT-6.1 Terra(max)OpenAI | 100% | epoch |
| 7 | Anthropic: Claude Opus 5.5(max)Anthropic | 100% | epoch |
| 8 | OpenAI: GPT-6.1 Sol(max)OpenAI | 100% | epoch |
| 9 | Anthropic: Claude Sonnet 5.5(max)Anthropic | 100% | epoch |
| 10 | OpenAI: GPT-6 Sol(max)OpenAI | 100% | epoch |
| 11 | OpenAI: GPT-6 Astra(max)OpenAI | 100% | epoch |
| 12 | gpt-5.5-pre-release_xhigh(xhigh)OpenAI | 100% | epoch |
| 13 | gpt-5.5-pro-pre-release_xhigh(xhigh)OpenAI | 100% | epoch |
| 14 | kimi-k3.1-pro-together_max(max)Moonshot | 100% | epoch |
| 15 | Google: Gemini 4 Carbon(high)Google DeepMind | 100% | epoch |
| 16 | Anthropic: Claude Fable 5.2(max)Anthropic | 99.7% | epoch |
| 17 | qwen3.8-max_xhigh(xhigh)Alibaba | 99.4% | epoch |
| 18 | muse-wonder-1.0_max(max)Meta AI | 99.4% | epoch |
| 19 | mistral-large-4-0-oss_high(high)Mistral AI | 99.4% | epoch |
| 20 | DeepSeek: DeepSeek V4.1 Flash 1017(max)DeepSeek | 99.4% | epoch |
| 21 | doubao-seed-evolving_high(high)ByteDance | 99.2% | epoch |
| 22 | Meta: Muse Spark 1.3(xhigh)Meta AI | 99.2% | epoch |
| 23 | SpaceXAI: Grok 4.6(xhigh)xAI | 99.2% | epoch |
| 24 | Google: Gemini 3.8 Flash(high)Google DeepMind | 98.9% | epoch |
| 25 | Anthropic: Claude Haiku 5.5(xhigh)Anthropic | 98.9% | epoch |
| 26 | OpenAI: GPT-6 Luna(max)OpenAI | 98.9% | epoch |
| 27 | DeepSeek: DeepSeek V4 Pro 0813(max)DeepSeek | 98.6% | epoch |
| 28 | DeepSeek: DeepSeek V4.1 Flash(max)DeepSeek | 98.3% | epoch |
| 29 | Mistral: Mistral Large 4(high)Mistral AI | 98.3% | epoch |
| 30 | SpaceXAI: Grok 4.7(xhigh)xAI | 98.1% | epoch |
| 31 | Meta: Muse Spark Open(xhigh)Meta AI | 98.1% | epoch |
| 32 | SpaceXAI: Grok 4.5(high)xAI | 97.8% | epoch |
| 33 | Anthropic: Claude Fable 5(low)Anthropic | 97.8% | epoch |
| 34 | Google: Gemini 3.7 Flash(high)Google DeepMind | 97.2% | epoch |
| 35 | doubao-seed-2.1-turbo-260628_high(high)ByteDance | 96.1% | epoch |
| 36 | MoonshotAI: Kimi K2.6Moonshot | 96.1% | epoch |
| 37 | doubao-seed-2.1-pro-260628_high(high)ByteDance | 95.6% | epoch |
| 38 | Google: Gemini 3 Flash Preview(high)Google DeepMind | 95.6% | epoch |
| 39 | Google: Gemini 3.1 Pro Preview(high)Google DeepMind | 95.6% | epoch |
| 40 | MoonshotAI: Kimi K2.7 CodeMoonshot | 95.6% | epoch |
| 41 | OpenAI: GPT-5.4(medium)OpenAI | 95.6% | epoch |
| 42 | Qwen: Qwen3.7 MaxAlibaba | 95.6% | epoch |
| 43 | claude-opus-4-6_64KAnthropic | 94.4% | epoch |
| 44 | DeepSeek: DeepSeek V4 Flash 0731(max)DeepSeek | 94.4% | epoch |
| 45 | doubao-seed-2.0-pro-260215_high(high)ByteDance | 93.9% | epoch |
| 46 | Z.ai: GLM 5.3 Flash(max)Z.ai (Zhipu AI) | 93.9% | epoch |
| 47 | Z.ai: GLM 5.1Z.ai (Zhipu AI) | 93.3% | epoch |
| 48 | Qwen: Qwen3.7 PlusAlibaba | 93.3% | epoch |
| 49 | SpaceXAI: Grok 4.3(high)xAI | 93.3% | epoch |
| 50 | Qwen: Qwen3.6 PlusAlibaba | 93.3% | epoch |
| 51 | Anthropic: Claude Opus 5(low)Anthropic | 93.3% | epoch |
| 52 | claude-opus-4-6_32KAnthropic | 93.1% | epoch |
| 53 | grok-4.20-0309-reasoningxAI | 92.2% | epoch |
| 54 | grok-4.20-0309-reasoningxAI | 92.2% | epoch |
| 55 | fireworks/kimi-k2p5Moonshot | 92.2% | epoch |
| 56 | fireworks/kimi-k2p5Moonshot | 92.2% | epoch |
| 57 | gemini-3-pro-previewGoogle DeepMind | 91.4% | epoch |
| 58 | gemini-3-pro-previewGoogle DeepMind | 91.4% | epoch |
| 59 | Anthropic: Claude Opus 4.6(max)Anthropic | 91.1% | epoch |
| 60 | qwen3.6-max-previewAlibaba | 91.1% | epoch |
| 61 | qwen3.6-max-previewAlibaba | 91.1% | epoch |
| 62 | doubao-seed-2.0-lite-260428_high(high)ByteDance | 91.1% | epoch |
| 63 | Z.ai: GLM 5.3(max)Z.ai (Zhipu AI) | 91.1% | epoch |
| 64 | Qwen: Qwen3.6 27BAlibaba | 91.1% | epoch |
| 65 | Thinking Machines: Inkling Small(xhigh)Thinking Machines | 90% | epoch |
| 66 | muse-sparkMeta AI | 88.9% | epoch |
| 67 | muse-sparkMeta AI | 88.9% | epoch |
| 68 | OpenAI: GPT-5.4 Mini(xhigh)OpenAI | 88.9% | epoch |
| 69 | Qwen: Qwen3.5 397B A17BAlibaba | 88.9% | epoch |
| 70 | Google: Gemini 3.5 Flash(low)Google DeepMind | 88.9% | epoch |
| 71 | Thinking Machines: Inkling(xhigh)Thinking Machines | 88.9% | epoch |
| 72 | OpenAI: gpt-oss-120b(high)OpenAI | 88.9% | epoch |
| 73 | deepseek-reasonerDeepSeek | 87.8% | epoch |
| 74 | deepseek-reasonerDeepSeek | 87.8% | epoch |
| 75 | qwen3.5-plusAlibaba | 86.7% | epoch |
| 76 | Qwen: Qwen3 235B A22B Thinking 2507Alibaba | 86.7% | epoch |
| 77 | Qwen: Qwen3.6 35B A3BAlibaba | 86.7% | epoch |
| 78 | Qwen: Qwen3.7 FlashAlibaba | 86.7% | epoch |
| 79 | Anthropic: Claude Opus 4.7(max)Anthropic | 86.7% | epoch |
| 80 | NVIDIA: Nemotron 3 UltraNvidia | 86.7% | epoch |
| 81 | claude-opus-4-5-20251101_32KAnthropic | 86.1% | epoch |
| 82 | claude-sonnet-4-6_32KAnthropic | 85.8% | epoch |
| 83 | Google: Gemini 2.5 ProGoogle DeepMind | 84.7% | epoch |
| 84 | Qwen: Qwen3.6 FlashAlibaba | 84.4% | epoch |
| 85 | OpenAI: o3(medium)OpenAI | 84.4% | epoch |
| 86 | qwen3.5-flashAlibaba | 84.4% | epoch |
| 87 | Anthropic: Claude Opus 4.8(none)Anthropic | 84.4% | epoch |
| 88 | grok-4-0709xAI | 84% | epoch |
| 89 | grok-4-0709xAI | 84% | epoch |
| 90 | Z.ai: GLM 4.7Z.ai (Zhipu AI) | 83.3% | epoch |
| 91 | kimi-k2-thinking-turboMoonshot | 83.1% | epoch |
| 92 | kimi-k2-thinking-turboMoonshot | 83.1% | epoch |
| 93 | Google: Gemma 4 26B A4B (minimal)Google DeepMind | 82.2% | epoch |
| 94 | claude-opus-4-5-20251101_16KAnthropic | 81.7% | epoch |
| 95 | Z.ai: GLM 5Z.ai (Zhipu AI) | 80% | epoch |
| 96 | Google: Gemini 3.6 Flash(minimal)Google DeepMind | 80% | epoch |
| 97 | Anthropic: Claude Sonnet 5(max)Anthropic | 80% | epoch |
| 98 | grok-3-mini-beta_high(high)xAI | 77.8% | epoch |
| 99 | claude-sonnet-4-5-20250929_32KAnthropic | 77.8% | epoch |
| 100 | claude-sonnet-4-5-20250929_59KAnthropic | 77.8% | epoch |
Run your own prompt on the top 3
Gemini 4 Argon, Qwen3.8 Max (0902), Claude Fable 5.1, side by side on the same prompt, with each one's time and cost. A free account includes starter credit.
Compare the leaders
More Mathematics benchmarks
MATH (Level 5)
The hardest tier of the MATH competition-problem benchmark.
FrontierMath
Research-level mathematics problems that take experts hours to days.
GSM8K
Grade-school word problems, long since saturated at the frontier.
LMArena: Math
Blind human preference between models on mathematical prompts.
Omni-MATH
Olympiad-level mathematical problem solving across a wide range of topics and difficulty levels.
MedCalc-Bench
Computing clinically relevant quantities from information in patient notes.
FrontierMath Tier 4 (v1, archived)
The original private set of exceptionally difficult research-mathematics problems.
FrontierMath Tier 4 (v2)
Corrected research-level mathematics at FrontierMath's hardest tier.
FrontierMath Tiers 1-3 (v2)
Corrected advanced mathematics from undergraduate depth through early research.
ProofBench
Writing formal mathematical proofs that pass Lean's verification checks.
AIME is maintained by MAA; curated by Epoch AI. Scores on this page are collected from openly-licensed sources and each links to its origin. See the full matrix to compare every model across every benchmark.