// BENCHMARK

AIME 2025 benchmark

AI model leaderboard for the AIME 2025 benchmark. Compare how large language models score on AIME 2025, see the full ranking, and understand what this AI benchmark measures. Kimi K2 Thinking currently leads with 100%. American Invitational Mathematics Examination 2025 — 30 hard high-school olympiad problems.

Leaderboard

# Model Organization Score Variant Source
#1 Kimi K2 Thinking Moonshot 100% thinking-heavy official ↗
#2 Tongyi DeepResearch Alibaba-NLP 100% with-search-and-python official ↗
#3 Kimi K2 Thinking Moonshot 99.1% thinking-w-python official ↗
#4 gpt-oss-20b OpenAI 98.7% high-with-tools official ↗
#5 gpt-oss-120b OpenAI 97.9% high-with-tools official ↗
#6 MAI-Thinking-1 Microsoft 97% official ↗
#7 GPT-5.5 Pro OpenAI 96.5% parallel-compute official ↗
#8 Gemini 3 Pro Google DeepMind 95% cited-deepseek-v3-2 official ↗
#9 GPT-5 OpenAI 94.6% thinking-high-cited-ring-1t official ↗
#10 GPT-5 OpenAI 94.6% cited-deepseek-v3-2 official ↗
#11 Kimi K2 Thinking Moonshot 94.5% thinking-cited-deepseek-v3-2 official ↗
#12 Kimi K2 Thinking Moonshot 94.5% thinking-no-tools official ↗
#13 Ring-1T Ant Group 93.4% thinking-avg@64 official ↗
#14 DeepSeek-V3.2 DeepSeek 93.1% thinking official ↗
#15 GPT-5.5 OpenAI 93% tools official ↗
#16 gpt-oss-120b OpenAI 92.5% high-no-tools official ↗
#17 gpt-oss-120b OpenAI 92.5% high reasoning, no tools official ↗
#18 Qwen3-235B-A22B-Thinking (Jul 2025) Qwen 92.3% thinking-cited-ring-1t official ↗
#19 GPT-5 OpenAI 92.15% High official ↗
#20 gpt-oss-20b OpenAI 91.7% high-no-tools official ↗
#21 gpt-oss-20b OpenAI 91.7% high reasoning, no tools official ↗
#22 DeepSeek V4-Pro DeepSeek 91.2% thinking official ↗
#23 Gemini 3.1 Pro Google DeepMind 89.7% official ↗
#24 Qwen3.6-Max-Preview Qwen 89.3% official ↗
#25 DeepSeek-V3.2-Exp DeepSeek 89.3% from-hf-readme official ↗
#26 DeepSeek-V3.1-Terminus DeepSeek 89.06% thinking-cited-ring-1t official ↗
#27 GPT-5.4 Thinking OpenAI 88.5% thinking official ↗
#28 gemini-2.5-pro Google DeepMind 88% cited-ring-1t official ↗
#29 Claude Opus 4.7 Anthropic 88% extended-thinking official ↗
#30 DeepSeek-R1-0528 DeepSeek 87.5% reasoning official ↗
#31 Claude Sonnet 4.5 Anthropic 87% cited-deepseek-v3-2 official ↗
#32 gemini-2.5-pro Google DeepMind 86.7% cited-phi4r official ↗
#33 o3 OpenAI 86.58% High official ↗
#34 gemini-2.5-pro Google DeepMind 85.75% official ↗
#35 DeepSeek V4-Flash DeepSeek 85.4% thinking official ↗
#36 EXAONE 4.0 (32B) LG AI Research 85.3% reasoning official ↗
#37 gpt-oss-120b OpenAI 84.59% official ↗
#38 Magistral Medium 1.1 Mistral AI 83.3% maj@64 official ↗
#39 Trinity-Large-Thinking Arcee AI 83.1% official ↗
#40 Kimi K2.6 Moonshot 82.9% official ↗
#41 Llama Nemotron Super 49B NVIDIA 82.71% reasoning-on-pass@1-avg@16 official ↗
#42 OpenReasoning-Nemotron-32B NVIDIA 82.71% official ↗
#43 GLM-5.1 Zhipu AI 82.7% official ↗
#44 o3-mini OpenAI 82.5% high-cited-phi4r official ↗
#45 DeepSeek-V3.1 DeepSeek 82.49% official ↗
#46 Claude Opus 4.6 Anthropic 82.1% extended-thinking official ↗
#47 Qwen3-235B-A22B Qwen 81.5% thinking official ↗
#48 Qwen3-235B-A22B-Thinking (Jul 2025) Qwen 81.5% reasoning official ↗
#49 K2 Think Mohamed bin Zayed University of Artificial Intelligence (MBZUAI) 81.24% avg@16 official ↗
#50 Qwen3-Max-Thinking Qwen 81.2% thinking official ↗
#51 GPT-5.5 Instant OpenAI 81.2% official ↗
#52 GPT-5.4 OpenAI 80.7% official ↗
#53 Qwen3.6-Plus Qwen 80.4% official ↗
#54 EXAONE Deep 32B LG AI Research 80% cons@64-from-hf-readme official ↗
#55 DeepSeek-R1 DeepSeek 80% cons@64-cited-exaone official ↗
#56 GLM-5 Zhipu AI 79.4% official ↗
#57 Muse Spark Meta AI 78.9% official ↗
#58 MiniMax-M2 MiniMax 78.3% cited-deepseek-v3-2 official ↗
#59 Phi-4-Reasoning-plus Microsoft 78% reasoning official ↗
#60 Phi-4-Reasoning-plus Microsoft 78% pass@1-avg@50 official ↗
#61 MiniMax-M1-80k MiniMax 76.9% 32 samples official ↗
#62 EXAONE Deep 7.8B LG AI Research 76.7% cons@64-from-hf-readme official ↗
#63 Magistral Small 1.1 Mistral AI 76.7% maj@64 official ↗
#64 QwQ-32B Qwen 76.7% cons@64-cited-exaone official ↗
#65 NVIDIA-Nemotron-Nano-12B-v2 NVIDIA 76.25% from-hf-readme official ↗
#66 Qwen3-235B-A22B Qwen 75.43% official ↗
#67 Kimi K2 Moonshot 75.2% k2-0905-w-python-cited-k2-thinking official ↗
#68 MiniMax-M1-40k MiniMax 74.6% 32 samples official ↗
#69 gpt-oss-20b OpenAI 74.58% official ↗
#70 NTele-R1-32B-V1 ZTE 74.49% from-hf-readme official ↗
#71 Qwen3-32B Qwen 73.33% thinking-cited-ntele-readme official ↗
#72 EXAONE Deep 2.4B LG AI Research 73.3% cons@64-from-hf-readme official ↗
#73 Qwen3-32B Qwen 72.9% thinking official ↗
#74 Qwen3-32B Qwen 72.9% reasoning official ↗
#75 Llama Nemotron Ultra 253B NVIDIA 72.5% reasoning-on-pass@1 official ↗
#76 NVIDIA-Nemotron-Nano-9B-v2 NVIDIA 72.1% from-hf-readme official ↗
#77 o1 OpenAI 71.4% cited-phi4r official ↗
#78 Qwen3-30B-A3B Qwen 70.9% thinking official ↗
#79 Ling-1T Ant Group 70.42% pass@1 official ↗
#80 DeepSeek-R1 DeepSeek 70.4% cited-phi4r official ↗
#81 Qwen3-14B Qwen 70.4% thinking official ↗
#82 gemini-2.5-pro Google DeepMind 70.1% lowthink-cited-ling-1t official ↗
#83 DeepSeek-R1 DeepSeek 70% cited-magistral official ↗
#84 QwQ-32B Qwen 69.17% official ↗
#85 Pangu Pro MoE Huawei 68.1% instruct-pass@1 official ↗
#86 QwQ-32B Qwen 67.3% cited-ntele-readme official ↗
#87 Qwen3-8B Qwen 67.3% thinking official ↗
#88 QwQ-32B Qwen 67.1% pass@1-cited-exaone official ↗
#89 DeepSeek-R1 DeepSeek 66.8% pass@1-cited-exaone official ↗
#90 DeepSeek-R1-Distill-Llama-70B DeepSeek 66.7% cons@64-cited-exaone official ↗
#91 EXAONE Deep 32B LG AI Research 65.8% pass@1-from-hf-readme official ↗
#92 QwQ-32B Qwen 65.8% cited-phi4r official ↗
#93 EXAONE Deep 32B LG AI Research 65.8% cited-phi4r official ↗
#94 Qwen3-4B Qwen 65.6% thinking official ↗
#95 DeepSeek-R1 DeepSeek 65.21% official ↗
#96 DeepSeek-R1 DeepSeek 65% cited-s1 official ↗
#97 Magistral Medium 1.1 Mistral AI 64.9% pass@1 official ↗
#98 Phi-4-Reasoning Microsoft 63.1% pass@1-avg@50 official ↗
#99 Magistral Small 1.1 Mistral AI 62.8% pass@1 official ↗
#100 LongCat-Flash Meituan Inc 61.25% avg@10 official ↗
#101 s1 Stanford University 60% s1-1-aime25-i-from-hf-readme official ↗
#102 EXAONE Deep 7.8B LG AI Research 59.6% pass@1-from-hf-readme official ↗
#103 GPT-5 OpenAI 59.43% main-cited-ling-1t official ↗
#104 Qwen3-30B-A3B Qwen 58.14% official ↗
#105 DeepSeek-R1-Distill-Qwen-7B Qwen 56.7% cons@64-cited-exaone official ↗
#106 DeepSeek-R1-Distill-Qwen-7B DeepSeek 56.7% cons@64-cited-exaone official ↗
#107 DeepSeek-V3.1-Terminus DeepSeek 55.21% non-thinking-cited-ling-1t official ↗
#108 DeepSeek-R1-Distill-Qwen-32B DeepSeek 55.21% cited-ntele-readme official ↗
#109 DeepSeek-R1-Distill-Qwen-32B Qwen 55.21% cited-ntele-readme official ↗
#110 o1-mini OpenAI 54.8% cited-phi4r official ↗
#111 DeepSeek-R1-Distill-Llama-70B DeepSeek 53.9% pass@1-cited-exaone official ↗
#112 DeepSeek-R1-Distill-Llama-8B DeepSeek 53.3% cons@64-cited-exaone official ↗
#113 Claude 3.7 Sonnet Anthropic 53% thinking-cited-phi4r official ↗
#114 DeepSeek-R1-Distill-Llama-70B DeepSeek 51.5% cited-phi4r official ↗
#115 Kimi K2 Moonshot 51% k2-0905-no-tools-cited-k2-thinking official ↗
#116 Kimi K2 Moonshot 50.16% instruct-0905-cited-ling-1t official ↗
#117 DeepSeek-V3-0324 DeepSeek 50% official ↗
#118 DeepSeek-R1-Distill-Qwen-14B Qwen 48.2% cited-mimo official ↗
#119 DeepSeek-R1-Distill-Qwen-14B DeepSeek 48.2% cited-mimo official ↗
#120 EXAONE Deep 2.4B LG AI Research 47.9% pass@1-from-hf-readme official ↗
#121 Llama Nemotron Nano 8B NVIDIA 47.1% reasoning-on-pass@1 official ↗
#122 DeepSeek-R1-Distill-Qwen-32B Qwen 46.1% cited-s1 official ↗
#123 DeepSeek-R1-Distill-Qwen-32B DeepSeek 46.1% cited-s1 official ↗
#124 EXAONE 4.0 (1.2B) LG AI Research 45.2% reasoning official ↗
#125 DeepSeek-R1-Distill-Qwen-7B Qwen 38.8% cited-mimo official ↗
#126 DeepSeek-R1-Distill-Qwen-7B DeepSeek 38.8% cited-mimo official ↗
#127 DeepSeek-R1-Distill-Qwen-7B DeepSeek 38.5% pass@1-cited-exaone official ↗
#128 DeepSeek-R1-Distill-Qwen-7B Qwen 38.5% pass@1-cited-exaone official ↗
#129 Qwen3-1.7B Qwen 36.8% thinking official ↗
#130 DeepSeek-R1-Distill-Qwen-1.5B DeepSeek 36.7% cons@64-cited-exaone official ↗
#131 DeepSeek-R1-Distill-Qwen-1.5B Qwen 36.7% cons@64-cited-exaone official ↗
#132 DeepSeek-R1-Distill-Llama-8B DeepSeek 33.6% pass@1-cited-exaone official ↗
#133 Mistral Medium 3 Mistral AI 30% maj@64-cited-magistral official ↗
#134 DeepSeek-V3 DeepSeek 28.8% cited-magistral official ↗
#135 s1 Stanford University 26.7% s1-aime25-i-from-hf-readme official ↗
#136 Qwen3-235B-A22B Qwen 24.7% non-thinking official ↗
#137 MiMo-7B-Base Xiaomi Corp 24.3% base official ↗
#138 DeepSeek-R1-Distill-Qwen-1.5B DeepSeek 23.9% pass@1-cited-exaone official ↗
#139 DeepSeek-R1-Distill-Qwen-1.5B Qwen 23.9% pass@1-cited-exaone official ↗
#140 Gemma 3 27B Google DeepMind 23.8% official ↗
#141 Qwen3-14B Qwen 23.3% non-thinking official ↗
#142 Qwen3-30B-A3B Qwen 21.6% non-thinking official ↗
#143 Kimi Linear Moonshot 21.3% sft-avg@64 official ↗
#144 Mistral Medium 3 Mistral AI 21.2% pass@1-cited-magistral official ↗
#145 Qwen3-8B Qwen 20.9% non-thinking official ↗
#146 Qwen3-32B Qwen 20.2% non-thinking official ↗
#147 Qwen3-4B Qwen 19.1% non-thinking official ↗
#148 Llama 4 Maverick Meta AI 18% official ↗
#149 Phi-4 Microsoft Research 17% baseline-cited-phi4r official ↗
#150 Llama Nemotron Ultra 253B NVIDIA 16.67% reasoning-off-pass@1 official ↗
#151 Qwen3-0.6B Qwen 15.1% thinking official ↗
#152 Llama 4 Scout Meta AI 10.2% pass@1-cited-pangu-pro-moe official ↗
#153 Qwen3-1.7B Qwen 9.8% non-thinking official ↗
#154 Qwen3-1.7B Qwen 9.8% non-reasoning official ↗
#155 Claude 3.5 Sonnet Anthropic 7.4% 1022-cited-mimo official ↗
#156 Qwen3-0.6B Qwen 2.6% non-reasoning official ↗
#157 Qwen3-0.6B Qwen 2.6% non-thinking official ↗
#158 Gemma 3 1B Google DeepMind 2.1% non-reasoning official ↗
#159 Llama Nemotron Nano 8B NVIDIA 0% reasoning-off-pass@1 official ↗

Frequently asked questions about AIME 2025

What is the AIME 2025 benchmark?

American Invitational Mathematics Examination 2025 — 30 hard high-school olympiad problems.

How is the AIME 2025 benchmark scored?

AIME 2025 is scored using the accuracy (%) metric, where a higher score is better. The maximum achievable score is 100.000. GenAIList aggregates reported scores from model providers and papers into a single ranked leaderboard.

Which AI model scores highest on AIME 2025?

As of the latest reported scores on GenAIList, Kimi K2 Thinking achieves the highest result on AIME 2025 with a score of 100%.

Is a higher AIME 2025 score better?

Yes. On AIME 2025 a higher score indicates better performance, so models near the top of the leaderboard are the strongest.