Mathematical Reasoning on GSM-8K (Accuracy)
97.3AccuracyFACT-E(Standard)
Evaluation Results
| Method | Links | |
|---|---|---|
| FACT-E(Standard)Backbone=DeepSeek-V32026.04 | 97.3 | |
| FACT-E(Lightweight)Backbone=DeepSeek-V32026.04 | 97.2 | |
| CONSISTENCYBackbone=DeepSeek-V32026.04 | 96.4 | |
| CoTBackbone=DeepSeek-V32026.04 | 96 | |
| REFLECTBackbone=DeepSeek-V32026.04 | 95.8 | |
| POLISHBackbone=DeepSeek-V32026.04 | 95.6 | |
| DENOISEBackbone=DeepSeek-V32026.04 | 95.4 | |
| FACT-EModel=DeepSeek-V3, K=52026.04 | 95 | |
| FACT-EModel=Qwen3, K=52026.04 | 94.8 | |
| BASEModel=DeepSeek-V3, K=52026.04 | 94.6 | |
| CONSIST.Model=DeepSeek-V3, K=52026.04 | 94.4 | |
| PaTaRMPolicy Model=Qwen3-8B2025.10 | 94.3 | |
| FACT-E(Lightweight)Backbone=Gpt-4o-mini2026.04 | 93.98 | |
| DENOISEModel=Qwen3, K=52026.04 | 93.8 | |
| REFLECTModel=DeepSeek-V3, K=52026.04 | 93.8 | |
| Skywork-Llama-3.1-8BPolicy Model=Qwen3-8B2025.10 | 93.7 | |
| DENOISEModel=DeepSeek-V3, K=52026.04 | 93.6 | |
| POLISHModel=Qwen3, K=52026.04 | 93.6 | |
| FACT-E(Standard)Backbone=Qwen3-14B2026.04 | 93.4 | |
| CoTBackbone=Qwen3-14B2026.04 | 93.2 | |
| DENOISEBackbone=Qwen3-14B2026.04 | 93.2 | |
| CONSISTENCYBackbone=Qwen3-14B2026.04 | 93.2 | |
| FACT-E(Lightweight)Backbone=Qwen3-14B2026.04 | 93.2 | |
| BASEModel=4o-mini, K=52026.04 | 93.2 | |
| BASEModel=Qwen3, K=52026.04 | 93.2 | |
| DENOISEModel=4o-mini, K=52026.04 | 93 | |
| CONSISTENCYBackbone=Gpt-4o-mini2026.04 | 92.8 | |
| REFLECTModel=Qwen3, K=52026.04 | 92.8 | |
| POLISHBackbone=Gpt-4o-mini2026.04 | 92.6 | |
| REFLECTBackbone=Qwen3-14B2026.04 | 92.6 | |
| CoTBackbone=Gpt-4o-mini2026.04 | 92.4 | |
| POLISHBackbone=Qwen3-14B2026.04 | 92.4 | |
| CONSIST.Model=Qwen3, K=52026.04 | 92.4 | |
| DENOISEBackbone=Gpt-4o-mini2026.04 | 92.2 | |
| FACT-EModel=4o-mini, K=52026.04 | 92.2 | |
| REFLECTBackbone=Gpt-4o-mini2026.04 | 92.1 | |
| CONSIST.Model=4o-mini, K=52026.04 | 92 | |
| FACT-E(Standard)Backbone=Gpt-4o-mini2026.04 | 91.8 | |
| POLISHModel=4o-mini, K=52026.04 | 91.6 | |
| POLISHModel=DeepSeek-V3, K=52026.04 | 91.6 | |
| REFLECTModel=4o-mini, K=52026.04 | 91 | |
| Rule-based RewardPolicy Model=Qwen3-8B2025.10 | 90.6 | |
| Base ModelPolicy Model=Qwen3-8B2025.10 | 89.8 | |
| LLAMA-4-SCOUTPARAMS=109B, Variant=FCSD2025.11 | 89.8 | |
| LLAMA-4-SCOUTPARAMS=109B, Variant=ORIGINAL2025.11 | 88.6 | |
| QWEN-3-30B MOEPARAMS=30B, Variant=FCSD2025.11 | 87.6 | |
| QWEN-3-30B MOEPARAMS=30B, Variant=ORIGINAL2025.11 | 86.9 | |
| FACT-E(Lightweight)Backbone=ChatGPT (Gpt-3.5-turbo)2026.04 | 81.4 | |
| PaTaRMPolicy Model=Qwen3-0.6B2025.10 | 81 | |
| CONSISTENCYBackbone=ChatGPT (Gpt-3.5-turbo)2026.04 | 80.6 | |
| LLAMA-4-SCOUTPARAMS=109B, Variant=SFT2025.11 | 78.6 | |
| REFLECTBackbone=ChatGPT (Gpt-3.5-turbo)2026.04 | 78.4 | |
| Rule-based RewardPolicy Model=Qwen3-0.6B2025.10 | 78.4 | |
| PiCaBackbone=Gemma-7B, #Params=10.22M, r=2562025.05 | 78.39 | |
| Full-FTBackbone=Gemma-7B, #Params=8.5B2025.05 | 78.09 | |
| CoTBackbone=ChatGPT (Gpt-3.5-turbo)2026.04 | 77.6 | |
| FACT-EModel=ChatGpt, K=52026.04 | 77.4 | |
| Skywork-Llama-3.1-8BPolicy Model=Qwen3-0.6B2025.10 | 77.3 | |
| POLISHBackbone=ChatGPT (Gpt-3.5-turbo)2026.04 | 77 | |
| FACT-E(Standard)Backbone=ChatGPT (Gpt-3.5-turbo)2026.04 | 77 | |
| DENOISEModel=ChatGpt, K=52026.04 | 77 | |
| Base ModelPolicy Model=Qwen3-0.6B2025.10 | 77 | |
| SVFT-RBackbone=Gemma-7B, #Params=19.8M, d=162025.05 | 76.81 | |
| DENOISEBackbone=ChatGPT (Gpt-3.5-turbo)2026.04 | 76.8 | |
| LoRABackbone=Gemma-7B, #Params=68.8M, r=322025.05 | 76.57 | |
| Full-FTBackbone=LLaMA-3-8B, #Params=8.0B2025.05 | 76.57 | |
| REFLECTModel=ChatGpt, K=52026.04 | 76.2 | |
| PiCaBackbone=LLaMA-3-8B, #Params=11.01M, r=2562025.05 | 76.12 | |
| QWEN-3-30B MOEPARAMS=30B, Variant=SFT2025.11 | 76 | |
| SVFT-RBackbone=LLaMA-3-8B, #Params=13.1M, d=122025.05 | 75.9 | |
| LoRABackbone=LLaMA-3-8B, #Params=56.6M, r=322025.05 | 75.89 | |
| DoRABackbone=LLaMA-3-8B, #Params=29.1M, r=162025.05 | 75.66 | |
| DoRABackbone=Gemma-7B, #Params=35.5M, r=162025.05 | 74.52 | |
| DoRABackbone=Gemma-7B, #Params=3.26M, r=12025.05 | 74.37 | |
| PiCaBackbone=Gemma-7B, #Params=0.64M, r=322025.05 | 74.3 | |
| BASEModel=ChatGpt, K=52026.04 | 74.2 | |
| PiCaBackbone=LLaMA-3-8B, #Params=1.38M, r=322025.05 | 73.54 | |
| SVFT-PBackbone=Gemma-7B, #Params=0.43M2025.05 | 73.5 | |
| LoRABackbone=Gemma-7B, #Params=0.82M, r=12025.05 | 72.4 | |
| POLISHModel=ChatGpt, K=52026.04 | 71.8 | |
| BOFTBackbone=Gemma-7B, #Params=2.90M, config=b=8, m=22025.05 | 71.79 | |
| CONSIST.Model=ChatGpt, K=52026.04 | 71.2 | |
| VeRABackbone=Gemma-7B, #Params=0.43M, r=10242025.05 | 71.11 | |
| DEEPSEEK-V2-LITEPARAMS=16B, Variant=ORIGINAL2025.11 | 70.1 | |
| SVFT-PBackbone=LLaMA-3-8B, #Params=0.48M2025.05 | 69.22 | |
| LoRABackbone=LLaMA-3-8B, #Params=1.77M, r=12025.05 | 68.84 | |
| DoRABackbone=LLaMA-3-8B, #Params=2.55M, r=12025.05 | 68.3 | |
| BOFTBackbone=LLaMA-3-8B, #Params=4.35M, config=b=8, m=22025.05 | 67.09 | |
| DEEPSEEK-V2-LITEPARAMS=16B, Variant=FCSD2025.11 | 64.2 | |
| VeRABackbone=LLaMA-3-8B, #Params=0.98M, r=10242025.05 | 63.76 | |
| DEEPSEEK-V2-LITEPARAMS=16B, Variant=SFT2025.11 | 54.3 | |
| PiCaBackbone=Gemma-2B, #Params=5.37M, r=2562025.05 | 52.77 | |
| Full-FTBackbone=Gemma-2B, #Params=2.5B2025.05 | 52.69 | |
| SVFT-RBackbone=Gemma-2B, #Params=6.35M, d=162025.05 | 50.03 | |
| DoRABackbone=Gemma-2B, #Params=13.5M, r=162025.05 | 44.27 | |
| LoRABackbone=Gemma-2B, #Params=26.2M, r=322025.05 | 43.06 | |
| DeepSeek V2-LiteAttention=MLA2026.04 | 41.47 | |
| PiCaBackbone=Gemma-2B, #Params=0.67M, r=322025.05 | 41.32 | |
| LCA2026.04 | 41.17 | |
| SVFT-PBackbone=Gemma-2B, #Params=0.19M2025.05 | 40.34 |