Mathematical Problem Solving on AIME (Acc, BAS, ECE, AURC)
85AccuracyG3 Mini
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| G3 MiniModel scale group=Mid-tier Models (10B to 70B params)2026.04 | 85 | 0.48 | 5.9 | 0.05 | |
| GPT-5 nanoScale=nano2026.05 | 81.7 | — | — | — | |
| gpt-oss-20bParameters=20B2026.05 | 79.2 | — | — | — | |
| GPT-o1Model scale group=Frontier / Large Models (> 100B params)2026.04 | 75 | 0.57 | 7.2 | 0.05 | |
| Qwen3-8BParameters=8B2026.05 | 74.2 | — | — | — | |
| Qwen3-4BParameters=4B2026.05 | 70.8 | — | — | — | |
| DS-R1Model scale group=Frontier / Large Models (> 100B params)2026.04 | 66.7 | 0.45 | 17.6 | 0.07 | |
| DS-V3.2Model scale group=Frontier / Large Models (> 100B params)2026.04 | 65 | 0.24 | 24.2 | 0.1 | |
| Grok-3Model scale group=Frontier / Large Models (> 100B params)2026.04 | 55 | 0.21 | 26.8 | 0.14 | |
| EngGPT2-16B-A3BParameters=16B, Active Parameters=3B, Architecture=MoE2026.05 | 49.2 | — | — | — | |
| Mist(L)Model scale group=Frontier / Large Models (> 100B params)2026.04 | 43.3 | -0.31 | 44 | 0.26 | |
| Mist(M)Model scale group=Mid-tier Models (10B to 70B params)2026.04 | 30 | -1.42 | 54.8 | 0.5 | |
| Ministral-3-8BParameters=8B2026.05 | 25 | — | — | — | |
| gemma-3-12b-itParameters=12B, Mode=instruct2026.05 | 20 | — | — | — | |
| Phi-4Model scale group=Small / Efficient Models (<10B params or highly optimized)2026.04 | 15 | -6.52 | 84.7 | 0.8 | |
| gemma-3-4b-itParameters=4B, Mode=instruct2026.05 | 12.5 | — | — | — | |
| GPT-4oModel scale group=Frontier / Large Models (> 100B params)2026.04 | 11.7 | -5.06 | 73.8 | 0.77 | |
| 4o-miniModel scale group=Small / Efficient Models (<10B params or highly optimized)2026.04 | 11.7 | -5.01 | 81.4 | 0.78 | |
| Llama 3.3Model scale group=Mid-tier Models (10B to 70B params)2026.04 | 8.3 | -3.81 | 52.8 | 0.82 | |
| Moonlight-16B-A3BParameters=16B, Active Parameters=3B, Architecture=MoE2026.05 | 5.8 | — | — | — | |
| Mist(S)Model scale group=Small / Efficient Models (<10B params or highly optimized)2026.04 | 5 | -1.68 | 62.5 | 0.88 | |
| Llama-3.2-3BParameters=3B, Version=3.22026.05 | 1.7 | — | — | — | |
| FastwebMIIA-7BParameters=7B2026.05 | 0 | — | — | — | |
| Minerva-7BParameters=7B2026.05 | 0 | — | — | — | |
| Velvet-14BParameters=14B2026.05 | 0 | — | — | — | |
| LLaMAntino-3-8BParameters=8B, Base Model=Llama 32026.05 | 0 | — | — | — | |
| Llama-3.1-8BParameters=8B, Version=3.12026.05 | 0 | — | — | — | |
| deepseek-moe-16bParameters=16B, Architecture=MoE2026.05 | 0 | — | — | — |