Math
AIME 2024/2025 Leaderboard
AIME (American Invitational Mathematics Examination) is a prestigious high-school competition of hard, integer-answer problems. It's a widely-cited yardstick for multi-step mathematical reasoning, here on the 2024–2025 papers.
Source: epoch86 open models ranked+197 proprietaryData through Oct 2026
Open models ranked on AIME 2024/2025
# shows rank among open models / rank overall (including proprietary).
| # | Model | Score | VRAM @ Q4 |
|---|---|---|---|
| 1 / 23 | DeepSeek V4 Pro 0813 · 1650.5B | 98.6% | ~991 GB |
| 2 / 25 | DeepSeek V4.1 Flash · 763.2B | 98.3% | ~458 GB |
| 3 / 29 | Muse Spark Open 816B A42B · 822.0B | 98.1% | ~543 GB |
| 4 / 39 | Kimi K3 · 2779.9B | 97.2% | ~1674 GB |
| 5 / 40 | DeepSeek V4 Pro · 1598.8B | 96.7% | ~960 GB |
| 6 / 42 | Kimi K2.6 · 1026.9B | 96.1% | ~620 GB |
| 7 / 50 | Kimi K2.7 Code · 1026.9B | 95.6% | ~620 GB |
| 8 / 55 | DeepSeek V4 Flash 0731 · 304.2B | 94.4% | ~183 GB |
| 9 / 57 | GLM 5.3 Flash · 321.3B | 93.9% | ~195 GB |
| 10 / 60 | GLM 5.1 · 753.9B | 93.3% | ~457 GB |
| 11 / 67 | Kimi K2.5 · 1026.9B | 92.2% | ~620 GB |
| 12 / 71 | GLM 5.3 · 753.3B | 91.1% | ~456 GB |
| 13 / 72 | Qwen3.6 27B · 27.8B | 91.1% | ~17 GB |
| 14 / 74 | Inkling Small · 266.0B | 90.0% | ~160 GB |
| 15 / 79 | GPT OSS 120B · 116.8B | 88.9% | ~71 GB |
| 16 / 80 | Inkling · 952.4B | 88.9% | ~572 GB |
| 17 / 81 | Qwen3.5 397B A17B · 403.4B | 88.9% | ~242 GB |
| 18 / 89 | NVIDIA Nemotron 3 Ultra 550B A55B BF16 · 560.5B | 86.7% | ~370 GB |
| 19 / 90 | Qwen3 235B A22B Thinking 2507 · 235.1B | 86.7% | ~142 GB |
| 20 / 92 | Qwen3.6 35B A3B · 36.0B | 86.7% | ~22 GB |
| 21 / 94 | GLM 5.2 · 753.3B | 86.4% | ~456 GB |
| 22 / 107 | GLM 4.7 · 358.3B | 83.3% | ~216 GB |
| 23 / 108 | Kimi K2 Thinking · 1026.4B | 83.1% | ~620 GB |
| 24 / 111 | Gemma 4 26B A4B IT · 25.8B | 82.2% | ~16 GB |
| 25 / 119 | GLM 5 · 753.9B | 80.0% | ~457 GB |
| 26 / 130 | Gemma 4 31B IT · 31.3B | 73.3% | ~20 GB |
| 27 / 140 | MiniMax M3 · 427.0B | 71.1% | ~257 GB |
| 28 / 142 | Qwen3 30B A3B Thinking 2507 · 30.5B | 70.3% | ~19 GB |
| 29 / 143 | Qwen3.5 35B A3B · 36.0B | 70.0% | ~22 GB |
| 30 / 149 | Seed OSS 36B Instruct · 36.2B | 67.5% | ~22 GB |
| 31 / 150 | Qwen3 32B · 32.8B | 66.9% | ~20 GB |
| 32 / 153 | DeepSeek R1 0528 · 684.5B | 66.4% | ~415 GB |
| 33 / 154 | Qwen3 14B · 14.8B | 66.4% | ~9.5 GB |
| 34 / 155 | GPT OSS 20B · 20.9B | 65.3% | ~13 GB |
| 35 / 160 | Qwen3 30B A3B · 30.5B | 62.8% | ~19 GB |
| 36 / 163 | Qwen3 30B A3B Instruct 2507 · 30.5B | 62.2% | ~19 GB |
| 37 / 164 | Qwen3.5 9B · 9.7B | 61.7% | ~6.4 GB |
| 38 / 168 | QwQ 32B · 32.8B | 59.2% | ~21 GB |
| 39 / 169 | GLM 4.7 Flash · 31.2B | 58.3% | ~20 GB |
| 40 / 175 | Qwen3 8B · 8.2B | 56.1% | ~5.5 GB |
| 41 / 176 | Qwen3.5 4B · 4.7B | 55.8% | ~3.3 GB |
| 42 / 177 | DeepSeek R1 Distill Qwen 32B · 32.8B | 55.6% | ~21 GB |
| 43 / 182 | DeepSeek R1 · 684.5B | 53.3% | ~415 GB |
| 44 / 185 | Qwen3 4B Instruct 2507 · 4.0B | 52.2% | ~2.9 GB |
| 45 / 186 | DeepSeek R1 Distill Llama 70B · 70.6B | 51.4% | ~43 GB |
| 46 / 188 | DeepSeek R1 Distill Qwen 14B · 14.8B | 50.6% | ~9.6 GB |
| 47 / 200 | DeepSeek R1 0528 Qwen3 8B · 8.2B | 43.9% | ~5.5 GB |
| 48 / 205 | DeepSeek v3 0324 · 684.5B | 37.8% | ~415 GB |
| 49 / 215 | Mistral Small 3.2 24B Instruct 2506 · 24.0B | 30.3% | ~15 GB |
| 50 / 216 | Magistral Small 2506 · 23.6B | 30.0% | ~15 GB |
| 51 / 219 | Magistral Small 2509 · 24.0B | 28.1% | ~15 GB |
| 52 / 222 | Gemma 3 27B IT · 27.4B | 22.5% | ~18 GB |
| 53 / 224 | DeepSeek R1 Distill Qwen 1.5B · 1.8B | 21.4% | ~1.4 GB |
| 54 / 225 | Llama 4 Maverick 17B 128E Instruct · 401.6B | 20.6% | ~265 GB |
| 55 / 227 | Gemma 3 12B IT · 12.2B | 16.7% | ~8.0 GB |
| 56 / 230 | DeepSeek v3 · 684.5B | 15.8% | ~415 GB |
| 57 / 231 | Phi 4 · 14.7B | 13.8% | ~9.5 GB |
| 58 / 233 | Llama 3.1 405B Instruct · 405.9B | 9.7% | ~268 GB |
| 59 / 235 | Mistral Large Instruct 2407 · 122.6B | 8.5% | ~81 GB |
| 60 / 236 | Qwen2.5 72B Instruct · 72.7B | 8.1% | ~45 GB |
| 61 / 237 | Qwen3 1.7B · 2.0B | 8.1% | ~1.8 GB |
| 62 / 238 | Llama 4 Scout 17B 16E Instruct · 108.6B | 7.8% | ~72 GB |
| 63 / 239 | Mistral Large Instruct 2411 · 122.6B | 7.8% | ~75 GB |
| 64 / 240 | Gemma 3 4B IT · 4.3B | 7.5% | ~2.8 GB |
| 65 / 241 | Qwen2.5 32B Instruct · 32.8B | 7.4% | ~21 GB |
| 66 / 245 | Mistral Small 24B Instruct 2501 · 23.6B | 6.7% | ~15 GB |
| 67 / 251 | Mistral Small 3.1 24B Instruct 2503 · 24.0B | 5.8% | ~15 GB |
| 68 / 252 | Llama 3.3 70B Instruct · 70.6B | 5.1% | ~47 GB |
| 69 / 255 | Llama 3.1 Tulu 3 70B DPO · 70.6B | 4.4% | ~43 GB |
| 70 / 257 | Meta Llama 3 70B Instruct · 70.6B | 4.3% | ~47 GB |
| 71 / 259 | Llama 3.1 70B Instruct · 70.6B | 3.6% | ~47 GB |
| 72 / 260 | Granite 4.0 Micro · 3.4B | 2.8% | ~2.5 GB |
| 73 / 261 | Llama 3.2 90B Vision Instruct · 88.6B | 2.6% | ~195 GB |
| 74 / 264 | Hermes 2 Theta Llama 3 70B · 70.6B | 2.5% | ~43 GB |
| 75 / 265 | Qwen2.5 7B Instruct · 7.6B | 2.5% | ~5.0 GB |
| 76 / 268 | Meta Llama 3 8B Instruct · 8.0B | 1.9% | ~5.3 GB |
| 77 / 271 | Llama 3.1 8B Instruct · 8.0B | 1.7% | ~5.3 GB |
| 78 / 272 | Gemma 2 27B IT · 27.2B | 1.4% | ~18 GB |
| 79 / 274 | Gemma 3 1B IT · 1000M | 1.1% | ~0.7 GB |
| 80 / 276 | Deepseek Llm 67B Chat · 67B | 0.8% | ~41 GB |
| 81 / 277 | Granite 4.0 1B · 1.6B | 0.8% | ~1.4 GB |
| 82 / 278 | Gemma 2 9B IT · 9.2B | 0.6% | ~6.1 GB |
| 83 / 280 | Llama 3.2 1B Instruct · 1.2B | 0.6% | ~0.8 GB |
| 84 / 281 | Granite 4.0 350M · 352M | 0.3% | ~0.6 GB |
| 85 / 282 | Mistral 7B Instruct v0.3 · 7.2B | 0.3% | ~4.9 GB |
| 86 / 283 | Llama 2 70B Chat HF · 69.0B | 0.0% | ~152 GB |
Score vs model size
Which models give the most quality for their size — the ones worth running locally.
- Granite 4.0 350M, 352M, score 0.3% — on the efficiency frontier (best score at its size or smaller).
- Gemma 3 1B IT, 1000M, score 1.1% — on the efficiency frontier (best score at its size or smaller).
- DeepSeek R1 Distill Qwen 1.5B, 2B, score 21.4% — on the efficiency frontier (best score at its size or smaller).
- Qwen3 4B Instruct 2507, 4B, score 52.2% — on the efficiency frontier (best score at its size or smaller).
- Qwen3.5 4B, 5B, score 55.8% — on the efficiency frontier (best score at its size or smaller).
- Qwen3 8B, 8B, score 56.1% — on the efficiency frontier (best score at its size or smaller).
- Qwen3.5 9B, 10B, score 61.7% — on the efficiency frontier (best score at its size or smaller).
- Qwen3 14B, 15B, score 66.4% — on the efficiency frontier (best score at its size or smaller).
- Gemma 4 26B A4B IT, 26B, score 82.2% — on the efficiency frontier (best score at its size or smaller).
- Qwen3.6 27B, 28B, score 91.1% — on the efficiency frontier (best score at its size or smaller).
- DeepSeek V4 Flash 0731, 304B, score 94.4% — on the efficiency frontier (best score at its size or smaller).
- DeepSeek V4.1 Flash, 763B, score 98.3% — on the efficiency frontier (best score at its size or smaller).
- DeepSeek V4 Pro 0813, 1.7T, score 98.6% — on the efficiency frontier (best score at its size or smaller).
AIME 2024/2025: frequently asked questions
- What is the best open LLM on AIME 2024/2025?
- DeepSeek V4 Pro 0813 is the top open model on AIME 2024/2025, scoring 98.6%. Among all models tested — including proprietary ones — it ranks #23. The top model overall is Claude Opus 5.5 Max (Anthropic) at 100.0%.
- What's the best AIME 2024/2025 model you can run on a 24 GB GPU?
- Qwen3.6 27B is the highest-scoring open model that fits in 24 GB at 4-bit quantization (~17 GB), scoring 91.1% on AIME 2024/2025.
- What's the best AIME 2024/2025 model you can run on a 12 GB GPU?
- Qwen3 14B is the highest-scoring open model that fits in 12 GB at 4-bit quantization (~9.5 GB), scoring 66.4% on AIME 2024/2025.
- Can open models match proprietary models on AIME 2024/2025?
- Not quite on AIME 2024/2025: the strongest proprietary model (Claude Opus 5.5 Max) scores 100.0%, ahead of the best open model (DeepSeek V4 Pro 0813) at 98.6% — but you can run the open one yourself.
Scores aggregated from epoch. llmrun does not run this benchmark — see the source for methodology, or the about benchmarks for what it measures.