Mathematical Reasoning on AIME 2025 (pass@k, cons@k, Pass Rate)
93Pass@kEAGER-INIT
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| EAGER-INITModel=GPT-Oss 20B2025.10 | 93 | 80 | 67 | |
| EAGERModel=GPT-Oss 20B2025.10 | 93 | 83 | 69 | |
| FULL PARALLELModel=GPT-Oss 20B2025.10 | 90 | 83 | 67 | |
| EAGERModel=Qwen3 4B2025.10 | 83 | 73 | 69 | |
| FULL PARALLELModel=Qwen3 4B2025.10 | 80 | 70 | 62 | |
| FULL PARALLELModel=DeepSeek 8B2025.10 | 80 | 67 | 65 | |
| EAGERModel=DeepSeek 8B2025.10 | 80 | 70 | 68 | |
| EAGER-INITModel=Qwen3 4B2025.10 | 77 | 70 | 61 | |
| EAGERModel=SmolLM 3B2025.10 | 73 | 33 | 31 | |
| EAGER-INITModel=DeepSeek 8B2025.10 | 73 | 60 | 59 | |
| FULL PARALLELModel=SmolLM 3B2025.10 | 53 | 0 | 6 | |
| EAGER-INITModel=SmolLM 3B2025.10 | 43 | 23 | 26 |