// BENCHMARK

AIME 2024 benchmark

AI model leaderboard for the AIME 2024 benchmark. Compare how large language models score on AIME 2024, see the full ranking, and understand what this AI benchmark measures. gpt-oss-120b currently leads with 96.6%. American Invitational Mathematics Examination 2024 — 30 olympiad problems.

Leaderboard

# Model Organization Score Variant Source
#1 gpt-oss-120b OpenAI 96.6% high-with-tools official ↗
#2 gpt-oss-20b OpenAI 96% high-with-tools official ↗
#3 gpt-oss-120b OpenAI 95.8% high reasoning, no tools official ↗
#4 gpt-oss-120b OpenAI 95.8% high-no-tools official ↗
#5 GPT-5 OpenAI 94.78% High official ↗
#6 Grok 4 xAI 94.3% official ↗
#7 Qwen3-235B-A22B-Thinking (Jul 2025) Qwen 94.1% official ↗
#8 o3 OpenAI 92.26% High official ↗
#9 gpt-oss-20b OpenAI 92.1% high reasoning, no tools official ↗
#10 gpt-oss-20b OpenAI 92.1% high-no-tools official ↗
#11 gemini-2.5-pro Google DeepMind 92% cited-phi4r official ↗
#12 DeepSeek-V3.1 DeepSeek 91.87% official ↗
#13 GLM 4.5 Zhipu AI 91% official ↗
#14 K2 Think Mohamed bin Zayed University of Artificial Intelligence (MBZUAI) 90.83% avg@16 official ↗
#15 o3 OpenAI 90.3% official ↗
#16 EXAONE Deep 32B LG AI Research 90% cons@64-from-hf-readme official ↗
#17 Magistral Medium 1.1 Mistral AI 90% maj@64 official ↗
#18 gpt-oss-120b OpenAI 89.58% official ↗
#19 GLM-4.5-Air Zhipu AI 89.4% official ↗
#20 DeepSeek-R1-0528 DeepSeek 89.3% official ↗
#21 gemini-2.5-pro Google DeepMind 88.7% official ↗
#22 o3-mini OpenAI 88% high-cited-phi4r official ↗
#23 Llama Nemotron Super 49B NVIDIA 87.5% reasoning-on-pass@1-avg@16 official ↗
#24 gemini-2.5-pro Google DeepMind 87.24% official ↗
#25 OpenReasoning-Nemotron-32B NVIDIA 87.09% official ↗
#26 DeepSeek-R1 DeepSeek 86.7% cons@64-cited-exaone official ↗
#27 QwQ-32B Qwen 86.7% cons@64-cited-exaone official ↗
#28 DeepSeek-R1-Distill-Llama-70B DeepSeek 86.7% cons@64-cited-exaone official ↗
#29 Qwen3-235B-A22B Qwen 86.68% official ↗
#30 MiniMax-M1-80k MiniMax 86% 32 samples official ↗
#31 Qwen3-235B-A22B Qwen 85.7% thinking official ↗
#32 Magistral Small 1.1 Mistral AI 83.3% maj@64 official ↗
#33 DeepSeek-R1-Distill-Qwen-7B Qwen 83.3% cons@64-cited-exaone official ↗
#34 DeepSeek-R1-Distill-Qwen-7B DeepSeek 83.3% cons@64-cited-exaone official ↗
#35 DeepSeek-R1-Distill-Qwen-32B Qwen 83.3% cons@64-cited-exaone official ↗
#36 DeepSeek-R1-Distill-Qwen-32B DeepSeek 83.3% cons@64-cited-exaone official ↗
#37 MiniMax-M1-40k MiniMax 83.3% 32 samples official ↗
#38 EXAONE Deep 7.8B LG AI Research 83.3% cons@64-from-hf-readme official ↗
#39 NTele-R1-32B-V1 ZTE 82.5% from-hf-readme official ↗
#40 Qwen3-32B Qwen 81.4% thinking official ↗
#41 Phi-4-Reasoning-plus Microsoft 81.3% pass@1-avg@5 official ↗
#42 Qwen3-30B-A3B Qwen 80.4% thinking official ↗
#43 Ling-1T Ant Group 80.21% pass@1 official ↗
#44 DeepSeek-R1-Distill-Llama-8B DeepSeek 80% cons@64-cited-exaone official ↗
#45 DeepSeek-R1 DeepSeek 79.8% cited-magistral official ↗
#46 QwQ-32B Qwen 79.5% pass@1-cited-exaone official ↗
#47 QwQ-32B Qwen 79.5% cited-phi4r official ↗
#48 QwQ-32B Qwen 79.38% official ↗
#49 Qwen3-14B Qwen 79.3% thinking official ↗
#50 Pangu Pro MoE Huawei 79.2% instruct-pass@1 official ↗
#51 Qwen3-32B Qwen 78.75% thinking-cited-ntele-readme official ↗
#52 DeepSeek-R1 DeepSeek 78.7% cited-phi4r official ↗
#53 gemini-2.5-pro Google DeepMind 77.5% lowthink-cited-ling-1t official ↗
#54 gpt-oss-20b OpenAI 76.88% reported in K2-Think official ↗
#55 EXAONE Deep 2.4B LG AI Research 76.7% cons@64-from-hf-readme official ↗
#56 QwQ-32B Qwen 76.25% cited-ntele-readme official ↗
#57 Qwen3-8B Qwen 76% thinking official ↗
#58 Claude Opus 4 Anthropic 75.7% official ↗
#59 o1 OpenAI 74.6% cited-phi4r official ↗
#60 Phi-4-Reasoning Microsoft 74.6% pass@1-avg@5 official ↗
#61 o1 OpenAI 74.4% cited-s1 official ↗
#62 DeepSeek-R1 DeepSeek 74.38% official ↗
#63 Qwen3-4B Qwen 73.8% thinking official ↗
#64 Magistral Medium 1.1 Mistral AI 73.6% pass@1 official ↗
#65 DeepSeek-R1-Distill-Qwen-32B DeepSeek 72.6% pass@1-cited-exaone official ↗
#66 DeepSeek-R1-Distill-Qwen-32B Qwen 72.6% pass@1-cited-exaone official ↗
#67 DeepSeek-R1-Distill-Qwen-32B DeepSeek 72.6% cited-s1 official ↗
#68 DeepSeek-R1-Distill-Qwen-32B Qwen 72.6% cited-s1 official ↗
#69 EXAONE Deep 32B LG AI Research 72.1% cited-phi4r official ↗
#70 EXAONE Deep 32B LG AI Research 72.1% pass@1-from-hf-readme official ↗
#71 DeepSeek-V3.1-Terminus DeepSeek 71.67% non-thinking-cited-ling-1t official ↗
#72 DeepSeek-R1-Zero DeepSeek 71% cited-magistral official ↗
#73 Magistral Small 1.1 Mistral AI 70.7% pass@1 official ↗
#74 Qwen3-30B-A3B Qwen 70.63% official ↗
#75 LongCat-Flash Meituan Inc 70.42% avg@10 official ↗
#76 EXAONE Deep 7.8B LG AI Research 70% pass@1-from-hf-readme official ↗
#77 DeepSeek-R1-Distill-Llama-70B DeepSeek 70% pass@1-cited-exaone official ↗
#78 DeepSeek-R1-Distill-Qwen-14B Qwen 69.7% cited-mimo official ↗
#79 DeepSeek-R1-Distill-Qwen-14B DeepSeek 69.7% cited-mimo official ↗
#80 DeepSeek-R1-Distill-Llama-70B DeepSeek 69.3% cited-phi4r official ↗
#81 GPT-5 OpenAI 67.6% main-cited-ling-1t official ↗
#82 Kimi K2 Moonshot 67.24% instruct-0905-cited-ling-1t official ↗
#83 DeepSeek-R1-Distill-Qwen-32B DeepSeek 64.17% cited-ntele-readme official ↗
#84 DeepSeek-R1-Distill-Qwen-32B Qwen 64.17% cited-ntele-readme official ↗
#85 o1-mini OpenAI 63.6% cited-phi4r official ↗
#86 OREAL 32B Shanghai AI Lab 60% from-hf-readme official ↗
#87 s1 Stanford University 56.7% s1-from-hf-readme official ↗
#88 s1 Stanford University 56.7% s1-1-from-hf-readme official ↗
#89 DeepSeek-R1-Distill-Qwen-7B Qwen 55.5% pass@1-cited-exaone official ↗
#90 DeepSeek-R1-Distill-Qwen-7B Qwen 55.5% cited-mimo official ↗
#91 DeepSeek-R1-Distill-Qwen-7B DeepSeek 55.5% cited-mimo official ↗
#92 DeepSeek-R1-Distill-Qwen-7B DeepSeek 55.5% pass@1-cited-exaone official ↗
#93 Claude 3.7 Sonnet Anthropic 55.3% thinking-cited-phi4r official ↗
#94 DeepSeek-R1-Distill-Qwen-1.5B Qwen 52.7% cons@64-cited-exaone official ↗
#95 DeepSeek-R1-Distill-Qwen-1.5B DeepSeek 52.7% cons@64-cited-exaone official ↗
#96 EXAONE Deep 2.4B LG AI Research 52.5% pass@1-from-hf-readme official ↗
#97 DeepSeek-R1-Distill-Llama-8B DeepSeek 50.4% pass@1-cited-exaone official ↗
#98 Qwen3-1.7B Qwen 48.3% thinking official ↗
#99 Mistral Medium 3 Mistral AI 43.4% maj@64-cited-magistral official ↗
#100 Qwen3-235B-A22B Qwen 40.1% non-thinking official ↗
#101 DeepSeek-V3 DeepSeek 39.2% cited-magistral official ↗
#102 DeepSeek-V3 DeepSeek 34% pass@1 official ↗
#103 OREAL 7B Shanghai AI Lab 33.3% from-hf-readme official ↗
#104 dots.llm1 Rednote 33.1% pass@1 official ↗
#105 MiMo-7B-Base Xiaomi Corp 32.9% base official ↗
#106 Qwen3-30B-A3B Qwen 32.8% non-thinking official ↗
#107 Qwen3-14B Qwen 31.7% non-thinking official ↗
#108 Qwen3-32B Qwen 31% non-thinking official ↗
#109 Qwen3-8B Qwen 29.1% non-thinking official ↗
#110 Llama 4 Scout Meta AI 29% pass@1-cited-pangu-pro-moe official ↗
#111 DeepSeek-R1-Distill-Qwen-1.5B Qwen 28.9% pass@1-cited-exaone official ↗
#112 DeepSeek-R1-Distill-Qwen-1.5B DeepSeek 28.9% pass@1-cited-exaone official ↗
#113 Mistral Medium 3 Mistral AI 26.8% pass@1-cited-magistral official ↗
#114 Eurus-2-7B-PRIME Tsinghua University 26.7% cited-oreal official ↗
#115 INTELLECT-MATH Prime Intellect 26.7% step-255-from-hf-readme official ↗
#116 Eurus-2-7B-PRIME Tsinghua University 26.7% from-hf-readme official ↗
#117 Qwen3-4B Qwen 25% non-thinking official ↗
#118 GPT-4o (Nov 2024) OpenAI 16.7% cited-oreal official ↗
#119 Claude 3.5 Sonnet Anthropic 16% 1022-cited-mimo official ↗
#120 Qwen3-1.7B Qwen 13.4% non-thinking official ↗
#121 Claude 3.5 Sonnet Anthropic 13.3% 1022-cited-oreal official ↗
#122 Qwen3-0.6B Qwen 10.7% thinking official ↗
#123 Qwen3-0.6B Qwen 3.4% non-thinking official ↗

Frequently asked questions about AIME 2024

What is the AIME 2024 benchmark?

American Invitational Mathematics Examination 2024 — 30 olympiad problems.

How is the AIME 2024 benchmark scored?

AIME 2024 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 2024?

As of the latest reported scores on GenAIList, gpt-oss-120b achieves the highest result on AIME 2024 with a score of 96.6%.

Is a higher AIME 2024 score better?

Yes. On AIME 2024 a higher score indicates better performance, so models near the top of the leaderboard are the strongest.