Math Reasoning on AIME25 (Accuracy)
94.6AIME25 AccuracyGPT-5
Evaluation results
| Method | Links | |
|---|---|---|
| GPT-5Evaluation Protocol=Closed-Source2026.01 | 94.6 | |
| Gemini2.5-ProEvaluation Protocol=Closed-Source2026.01 | 86.7 | |
| DiffMASBackbone=Qwen3-14B2026.04 | 66.7 | |
| LatentMASBackbone=Qwen3-14B2026.04 | 63.3 | |
| TextMASBackbone=Qwen3-14B2026.04 | 60 | |
| SingleBackbone=Qwen3-14B2026.04 | 56.7 | |
| SingleBackbone=DeepSeek-R1-Distill-Qwen-32B2026.04 | 50 | |
| TextMASBackbone=DeepSeek-R1-Distill-Qwen-32B2026.04 | 50 | |
| DiffMASBackbone=DeepSeek-R1-Distill-Qwen-32B2026.04 | 50 | |
| LatentMASBackbone=DeepSeek-R1-Distill-Qwen-32B2026.04 | 43.3 | |
| ATLAS (cluster)Evaluation Protocol=In-Distribution2026.01 | 40 | |
| DiffMASBackbone=Mistral3-8B2026.04 | 40 | |
| GPT-4.1Evaluation Protocol=Closed-Source2026.01 | 33.3 | |
| ATLAS (RL)Evaluation Protocol=Out-of-Distribution2026.01 | 33.3 | |
| LatentMASBackbone=Mistral3-8B2026.04 | 33.3 | |
| RouterDCEvaluation Protocol=In-Distribution2026.01 | 23.3 | |
| SingleBackbone=Mistral3-8B2026.04 | 23.3 | |
| TextMASBackbone=Mistral3-8B2026.04 | 23.3 | |
| FS RouterEvaluation Protocol=Training-free, Prompting Strategy=Few-shot2026.01 | 13.3 | |
| BertRouterEvaluation Protocol=In-Distribution2026.01 | 13.3 | |
| MLPRouterEvaluation Protocol=In-Distribution2026.01 | 10 | |
| GPT-4oEvaluation Protocol=Closed-Source2026.01 | 6.7 | |
| ZS RouterEvaluation Protocol=Training-free, Prompting Strategy=Zero-shot2026.01 | 6.7 | |
| BertRouterEvaluation Protocol=Out-of-Distribution2026.01 | 6.7 | |
| Random RouterEvaluation Protocol=Training-free2026.01 | 3.3 | |
| RouterDCEvaluation Protocol=Out-of-Distribution2026.01 | 3.3 | |
| MLPRouterEvaluation Protocol=Out-of-Distribution2026.01 | 3.3 | |
| ATLAS (cluster)Evaluation Protocol=Out-of-Distribution2026.01 | 3.3 |