Mathematical Reasoning on GSM-Infinite Hard
50.4AccuracyDeepSeek-V3.2
Evaluation Results
| Method | Links | |
|---|---|---|
| DeepSeek-V3.2Length=16K, #Activated Params=37B, #Total Params=671B2026.01 | 50.4 | |
| DeepSeek-V3.2Length=32K, #Activated Params=37B, #Total Params=671B2026.01 | 45.2 | |
| DeepSeek-V3.1Length=16K, #Activated Params=37B, #Total Params=671B2026.01 | 41.5 | |
| DeepSeek-V3.1Length=32K, #Activated Params=37B, #Total Params=671B2026.01 | 38.8 | |
| MiMo-V2-FlashLength=16K, #Activated Params=15B, #Total Params=309B2026.01 | 37.7 | |
| DeepSeek-V3.1Length=64K, #Activated Params=37B, #Total Params=671B2026.01 | 34.7 | |
| Kimi-K2Length=16K, #Activated Params=32B, #Total Params=1043B2026.01 | 34.6 | |
| MiMo-V2-FlashLength=32K, #Activated Params=15B, #Total Params=309B2026.01 | 33.7 | |
| DeepSeek-V3.2Length=64K, #Activated Params=37B, #Total Params=671B2026.01 | 32.6 | |
| MiMo-V2-FlashLength=64K, #Activated Params=15B, #Total Params=309B2026.01 | 31.5 | |
| MiMo-V2-FlashLength=128K, #Activated Params=15B, #Total Params=309B2026.01 | 29 | |
| DeepSeek-V3.1Length=128K, #Activated Params=37B, #Total Params=671B2026.01 | 28.7 | |
| Kimi-K2Length=32K, #Activated Params=32B, #Total Params=1043B2026.01 | 26.1 | |
| DeepSeek-V3.2Length=128K, #Activated Params=37B, #Total Params=671B2026.01 | 25.7 | |
| Kimi-K2Length=64K, #Activated Params=32B, #Total Params=1043B2026.01 | 16 | |
| Kimi-K2Length=128K, #Activated Params=32B, #Total Params=1043B2026.01 | 8.8 |