Reasoning on ARC
94.5AccuracyQwen-7B-Instruct
Evaluation results
| Method | Links | |||||||
|---|---|---|---|---|---|---|---|---|
| Qwen-7B-InstructRouter=false2025.10 | 94.5 | — | — | — | — | 112 | — | |
| Qwen-7B-InstructRouter=true2025.10 | 94.5 | — | — | — | — | 111.8 | — | |
| Qwen3-8BLoss function=base+aux2026.02 | 92.34 | — | — | — | — | — | — | |
| Qwen3-8BLoss function=base2026.02 | 91.98 | — | — | — | — | — | — | |
| Qwen3-4BLoss function=base+aux2026.02 | 90.84 | — | — | — | — | — | — | |
| HARP-EModel=QWEN3-8B, Train Dataset=Tulu2026.06 | 90.8 | — | — | — | — | — | — | |
| Qwen3-4BLoss function=base2026.02 | 89.98 | — | — | — | — | — | — | |
| LLaMA-8B-InstructRouter=true2025.10 | 89.4 | — | — | — | — | 104 | — | |
| FLBackbone=PHI-4-14B, SFT Dataset=UF2026.04 | 89 | — | — | — | — | — | — | |
| GEMBackbone=PHI-4-14B, SFT Dataset=UF2026.04 | 88.9 | — | — | — | — | — | — | |
| CEBackbone=PHI-4-14B, SFT Dataset=UF2026.04 | 88.7 | — | — | — | — | — | — | |
| TOFUBackbone=PHI-4-14B, SFT Dataset=UF2026.04 | 88.7 | — | — | — | — | — | — | |
| LLaMA-8B-InstructRouter=false2025.10 | 88.5 | — | — | — | — | 106.9 | — | |
| PPCVBackbone=Llama-3.1-8B-Instruct2026.02 | 88.31 | — | — | — | — | — | — | |
| FLBackbone=QWEN-3-8B, SFT Dataset=UF2026.04 | 88.3 | — | — | — | — | — | — | |
| TOFUBackbone=QWEN-3-8B, SFT Dataset=UF2026.04 | 88.2 | — | — | — | — | — | — | |
| CEBackbone=QWEN-3-8B, SFT Dataset=UF2026.04 | 88.1 | — | — | — | — | — | — | |
| GEMBackbone=QWEN-3-8B, SFT Dataset=UF2026.04 | 87.9 | — | — | — | — | — | — | |
| SFT+GRPOBackbone=Qwen2.5-7B2026.06 | 87.8 | — | — | — | — | — | 405 | |
| ADaPTthinkBackbone=Qwen2.5-7B2026.06 | 87.7 | — | — | — | — | — | 327 | |
| FDM-ALLDM Model=LLaDA-1.52025.12 | 87.66 | — | 38.43 | — | — | — | — | |
| GEMBackbone=QWEN-3-8B, SFT Dataset=ALPACA2026.04 | 87 | — | — | — | — | — | — | |
| CEBackbone=QWEN-3-8B, SFT Dataset=ALPACA2026.04 | 86.9 | — | — | — | — | — | — | |
| FLBackbone=QWEN-3-8B, SFT Dataset=ALPACA2026.04 | 86.9 | — | — | — | — | — | — | |
| TOFUBackbone=QWEN-3-8B, SFT Dataset=ALPACA2026.04 | 86.9 | — | — | — | — | — | — | |
| ADaPTBackbone=Qwen2.5-7B2026.06 | 86.9 | — | — | — | — | — | 78 | |
| λ-PRBackbone=PHI-4-14B, SFT Dataset=ALPACA2026.04 | 86.7 | — | — | — | — | — | — | |
| ARMBackbone=Qwen2.5-7B2026.06 | 86.7 | — | — | — | — | — | 204 | |
| ProbabilityLLDM Model=LLaDA-1.5, Threshold (T)=1282025.12 | 86.38 | — | 21.71 | — | — | — | — | |
| FDM-ALLDM Model=LLaDA2025.12 | 86.3 | — | 38.2 | — | — | — | — | |
| GEMBackbone=PHI-4-14B, SFT Dataset=ALPACA2026.04 | 86.2 | — | — | — | — | — | — | |
| CEBackbone=PHI-4-14B, SFT Dataset=ALPACA2026.04 | 86.1 | — | — | — | — | — | — | |
| λ-PRBackbone=QWEN-3-8B, SFT Dataset=ALPACA2026.04 | 86.1 | — | — | — | — | — | — | |
| MarginLLDM Model=LLaDA-1.5, Threshold (T)=1282025.12 | 85.91 | — | 21.34 | — | — | — | — | |
| TOFUBackbone=PHI-4-14B, SFT Dataset=ALPACA2026.04 | 85.8 | — | — | — | — | — | — | |
| FLBackbone=PHI-4-14B, SFT Dataset=ALPACA2026.04 | 85.7 | — | — | — | — | — | — | |
| HARP-CModel=QWEN3-4B, Train Dataset=SI2026.06 | 85.7 | — | — | — | — | — | — | |
| EBLLDM Model=LLaDA-1.52025.12 | 85.68 | — | 34.89 | — | — | — | — | |
| AdaPreLoRA AdamWModel=Qwen2-7B, Rank=82026.05 | 85.6 | — | — | — | — | — | — | |
| Phi-DecodingBackbone=Llama-3.1-8B-Instruct2026.02 | 85.41 | — | — | — | — | — | — | |
| WINOLLDM Model=LLaDA-1.52025.12 | 85.31 | — | 35.27 | — | — | — | — | |
| Scaled AdamWModel=Qwen2-7B, Rank=82026.05 | 85.3 | — | — | — | — | — | — | |
| ADaPTanswerBackbone=Qwen2.5-7B2026.06 | 85.3 | — | — | — | — | — | 42 | |
| R-4BBackbone=Qwen2.5-7B2026.06 | 84.7 | — | — | — | — | — | 274 | |
| EntropyLLDM Model=LLaDA-1.5, Threshold (T)=1282025.12 | 84.58 | — | 20.48 | — | — | — | — | |
| Predictive DecodingBackbone=Llama-3.1-8B-Instruct2026.02 | 84.56 | — | — | — | — | — | — | |
| MarginLLDM Model=LLaDA, Threshold (T)=1282025.12 | 84.34 | — | 20.65 | — | — | — | — | |
| AdamWModel=Qwen2-7B, Rank=82026.05 | 84.3 | — | — | — | — | — | — | |
| CESBackbone=DeepSeek-R1-Distill-1.5B2026.05 | 83.6 | — | — | — | — | — | 542 | |
| FDM-ALLDM Model=LLaDA-MoE2025.12 | 83.33 | — | 12.62 | — | — | — | — | |
| TLMREBackbone=Qwen2.5-7B2026.06 | 82.6 | — | — | — | — | — | 287 | |
| ProbabilityLLDM Model=LLaDA, Threshold (T)=1282025.12 | 82.16 | — | 21.24 | — | — | — | — | |
| SHED-OModel=QWEN3-4B, Train Dataset=SI2026.06 | 81.9 | — | — | — | — | — | — | |
| Guided DecodingBackbone=Llama-3.1-8B-Instruct2026.02 | 81.74 | — | — | — | — | — | — | |
| LoRA-Pro AdamWModel=Qwen2-7B, Rank=82026.05 | 80.9 | — | — | — | — | — | — | |
| SOAPModel=Qwen2-7B, Rank=82026.05 | 80.9 | — | — | — | — | — | — | |
| Tree-of-ThoughtBackbone=Llama-3.1-8B-Instruct2026.02 | 80.72 | — | — | — | — | — | — | |
| LLaMA3.1-8BLoss function=base+aux2026.02 | 80.6 | — | — | — | — | — | — | |
| LLaMA3.1-8BLoss function=base2026.02 | 80.41 | — | — | — | — | — | — | |
| DAPOBackbone=DeepSeek-R1-Distill-1.5B2026.05 | 79.7 | — | — | — | — | — | 543 | |
| WINOLLDM Model=LLaDA2025.12 | 79.44 | — | 34.17 | — | — | — | — | |
| λ-PRBackbone=QWEN-3-8B, SFT Dataset=UF2026.04 | 78.8 | — | — | — | — | — | — | |
| GEMBackbone=Mistral-12B, SFT Dataset=UF2026.04 | 77.3 | — | — | — | — | — | — | |
| EntropyLLDM Model=LLaDA, Threshold (T)=1282025.12 | 77.2 | — | 19.91 | — | — | — | — | |
| λ-PRBackbone=PHI-4-14B, SFT Dataset=UF2026.04 | 77.2 | — | — | — | — | — | — | |
| FLBackbone=Mistral-12B, SFT Dataset=ALPACA2026.04 | 76.8 | — | — | — | — | — | — | |
| GEMBackbone=Mistral-12B, SFT Dataset=ALPACA2026.04 | 76.8 | — | — | — | — | — | — | |
| CEBackbone=Mistral-12B, SFT Dataset=UF2026.04 | 76.3 | — | — | — | — | — | — | |
| TOFUBackbone=Mistral-12B, SFT Dataset=UF2026.04 | 76.1 | — | — | — | — | — | — | |
| MarginLLDM Model=LLaDA-MoE, Threshold (T)=1282025.12 | 75.48 | — | 8.56 | — | — | — | — | |
| CEBackbone=Mistral-12B, SFT Dataset=ALPACA2026.04 | 74.6 | — | — | — | — | — | — | |
| TOFUBackbone=OLMO-2-13B, SFT Dataset=ALPACA2026.04 | 74.6 | — | — | — | — | — | — | |
| LLaMA-3B-InstructRouter=true2025.10 | 74.5 | — | — | — | — | 99.5 | — | |
| CEBackbone=OLMO-2-13B, SFT Dataset=UF2026.04 | 74.4 | — | — | — | — | — | — | |
| FLBackbone=Mistral-12B, SFT Dataset=UF2026.04 | 74.3 | — | — | — | — | — | — | |
| GEMBackbone=OLMO-2-13B, SFT Dataset=ALPACA2026.04 | 74.2 | — | — | — | — | — | — | |
| GEMBackbone=OLMO-2-13B, SFT Dataset=UF2026.04 | 74.1 | — | — | — | — | — | — | |
| TOFUBackbone=OLMO-2-13B, SFT Dataset=UF2026.04 | 74.1 | — | — | — | — | — | — | |
| FLBackbone=OLMO-2-13B, SFT Dataset=ALPACA2026.04 | 74 | — | — | — | — | — | — | |
| HARP-CModel=LLAMA-8B, Train Dataset=SI2026.06 | 74 | — | — | — | — | — | — | |
| HARP-EModel=LLAMA-8B, Train Dataset=SI2026.06 | 74 | — | — | — | — | — | — | |
| TOFUBackbone=Mistral-12B, SFT Dataset=ALPACA2026.04 | 73.9 | — | — | — | — | — | — | |
| FLBackbone=OLMO-2-13B, SFT Dataset=UF2026.04 | 73.7 | — | — | — | — | — | — | |
| EBLLDM Model=LLaDA2025.12 | 73.55 | — | 32.01 | — | — | — | — | |
| CEBackbone=OLMO-2-13B, SFT Dataset=ALPACA2026.04 | 73.5 | — | — | — | — | — | — | |
| LLaMA-3B-InstructRouter=false2025.10 | 73.5 | — | — | — | — | 103.9 | — | |
| SFTBackbone=Qwen2.5-7B2026.06 | 73.2 | — | — | — | — | — | 949 | |
| ProbabilityLLDM Model=LLaDA-MoE, Threshold (T)=1282025.12 | 72.74 | — | 8.68 | — | — | — | — | |
| TOFUBackbone=LLAMA-3.1-8B, SFT Dataset=ALPACA2026.04 | 72.5 | — | — | — | — | — | — | |
| TOFUBackbone=LLAMA-3.1-8B, SFT Dataset=UF2026.04 | 72.3 | — | — | — | — | — | — | |
| GEMBackbone=LLAMA-3.1-8B, SFT Dataset=ALPACA2026.04 | 72.1 | — | — | — | — | — | — | |
| SFT+GRPOBackbone=Qwen2.5-3B2026.06 | 72.1 | — | — | — | — | — | 577 | |
| CEBackbone=LLAMA-3.1-8B, SFT Dataset=UF2026.04 | 71.8 | — | — | — | — | — | — | |
| GEMBackbone=LLAMA-3.1-8B, SFT Dataset=UF2026.04 | 71.8 | — | — | — | — | — | — | |
| Apertus-8Bmode=Zero-shot, model_type=Base2026.05 | 71.66 | — | — | — | — | — | — | |
| EBLLDM Model=LLaDA-MoE2025.12 | 71.5 | — | 8.98 | — | — | — | — | |
| ADaPTBackbone=Qwen2.5-3B2026.06 | 71.5 | — | — | — | — | — | 74 | |
| CEBackbone=LLAMA-3.1-8B, SFT Dataset=ALPACA2026.04 | 71.4 | — | — | — | — | — | — | |
| FLBackbone=LLAMA-3.1-8B, SFT Dataset=ALPACA2026.04 | 71.2 | — | — | — | — | — | — | |
| FLBackbone=LLAMA-3.1-8B, SFT Dataset=UF2026.04 | 71.1 | — | — | — | — | — | — |