Mathematical Reasoning on MathBench
95.6AccuracyI-DLM
Evaluation Results
| Method | Links | |
|---|---|---|
| I-DLMParams=32B, Decoding=ISD (N=4, sampling)2026.04 | 95.6 | |
| Qwen3Params=32B2026.04 | 95.5 | |
| Qwen3Params=8B2026.04 | 93.1 | |
| I-DLMParams=8B, Decoding=ISD (N=4, sampling)2026.04 | 89.1 | |
| LLaDA-2.1-miniParams=16B2026.04 | 84.2 | |
| SDARParams=30B2026.04 | 79.3 | |
| SDARParams=8B2026.04 | 76.9 | |
| TaskFT-SPEED-24+BoSFine-tuning=Nemotron-Math, K=24, Visibility Policy=SPEED+BoS2026.05 | 53 | |
| TaskFT-SPEED-20+BoSFine-tuning=Nemotron-Math, K=20, Visibility Policy=SPEED+BoS2026.05 | 52.8 | |
| TaskFT-FullFine-tuning=Nemotron-Math, Depth=Full2026.05 | 50.4 | |
| TaskFT-SPEED-28+BoSFine-tuning=Nemotron-Math, K=28, Visibility Policy=SPEED+BoS2026.05 | 48.8 |