List your company
ResearchFormerly WizWand

Math Score on GSM8K

94.01GSM8K Math ScoreQwen3.5-9B + AR-SFT
23.945242.135160.32578.5149May 25, 2026May 28, 2026May 31, 2026Jun 4, 2026Jun 7, 2026Jun 10, 2026Jun 14, 2026
Updated 4mo ago

Evaluation results

MethodLinks
Qwen3.5-9B + AR-SFTModel Size=9B, Training Protocol=AR-SFT, Decoding Strategy=Native AR decoding2026.0694.01
FLARE-9BModel Size=9B, Training Protocol=FLARE, Decoding Strategy=AR-Trust sampling2026.0693.33
Qwen3.5-4B + AR-SFTModel Size=4B, Training Protocol=AR-SFT, Decoding Strategy=Native AR decoding2026.0691.66
FLARE-4BModel Size=4B, Training Protocol=FLARE, Decoding Strategy=AR-Trust sampling2026.0691.05
(0,2,2)Model=Qwen3, Pruning Ratio=50%2026.0689.2
Qwen3.5-4B (released)Model Size=4B, Training Protocol=Released, Decoding Strategy=Native AR decoding2026.0689.16
Qwen3.5-9B (released)Model Size=9B, Training Protocol=Released, Decoding Strategy=Native AR decoding2026.0689.16
FLARE-2BModel Size=2B, Training Protocol=FLARE, Decoding Strategy=AR-Trust sampling2026.0684.46
Qwen3.5-2B + AR-SFTModel Size=2B, Training Protocol=AR-SFT, Decoding Strategy=Native AR decoding2026.0683.93
MANModel=ERNIE, Pruning Ratio=50%2026.0679
Qwen3.5-2B (released)Model Size=2B, Training Protocol=Released, Decoding Strategy=Native AR decoding2026.0677.63
DSRMidtraining Schedule=DSR2026.0569.82
FullModel=OLMoE, Pruning Ratio=0%2026.0668.2
Unified 15B MoEMidtraining Corpus=1k→8k2026.0566.74
WSDMidtraining Schedule=WSD2026.0565.59
Cosine-decayMidtraining Schedule=Cos2026.0562.97
Unified 15B MoEMidtraining Corpus=4k→8k2026.0561.89
Unified 15B MoEMidtraining Corpus=4k2026.0561.74
FullModel=DeepSeek, Pruning Ratio=0%2026.0661
Unified 15B MoEPretraining Corpus=4k→8k2026.0529.72
Unified 15B MoEPretraining Corpus=4k2026.0527.56
Unified 15B MoEPretraining Corpus=1k→8k2026.0526.64