General Reasoning on MMLU (Accuracy and Token Cost)
86.21AccuracyPivot-SFT
Evaluation results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| Pivot-SFTModel Size=14B, GPU Hour=1202025.05 | 86.21 | — | — | — | — | — | |
| InfiFusionModel Size=14B, GPU Hour=1602025.05 | 85.81 | — | — | — | — | — | |
| Phi-4Model Size=14B, GPU Hour=∼1.0M2025.05 | 85.62 | — | — | — | — | — | |
| InfiGFusionModel Size=14B, GPU Hour=1952025.05 | 85.24 | — | — | — | — | — | |
| MiniLogitModel Size=14B, GPU Hour=2202025.05 | 84.58 | — | — | — | — | — | |
| FuseChatModel Size=14B, GPU Hour=6502025.05 | 84.23 | — | — | — | — | — | |
| FuseLLMModel Size=14B, GPU Hour=2252025.05 | 83.92 | — | — | — | — | — | |
| Mistral-SmallModel Size=24B, GPU Hour=∼1.6M2025.05 | 81.69 | — | — | — | — | — | |
| DenserBackbone=Qwen3-14B2025.12 | 81.5 | -58.7 | — | — | — | — | |
| Self-VerificationBackbone=Qwen3-14B2025.12 | 80.4 | 142.8 | — | — | — | — | |
| Qwen2.5-InstructModel Size=14B, GPU Hour=∼1.8M2025.05 | 80.22 | — | — | — | — | — | |
| Process SupervisionBackbone=Qwen3-14B2025.12 | 80.1 | 156.3 | — | — | — | — | |
| Reflection-CoTBackbone=Qwen3-14B2025.12 | 79.5 | 127.6 | — | — | — | — | |
| TokenSkip (0.9 ratio)Backbone=Qwen3-14B2025.12 | 79.4 | -53.5 | — | — | — | — | |
| Think-to-ThinkBackbone=Qwen3-14B2025.12 | 79.1 | 187.2 | — | — | — | — | |
| Self-ConsistencyBackbone=Qwen3-14B2025.12 | 78.5 | 289.4 | — | — | — | — | |
| Tree-of-ThoughtBackbone=Qwen3-14B2025.12 | 78.2 | 345.7 | — | — | — | — | |
| Chain-of-Thought (CoT)Backbone=Qwen3-14B2025.12 | 77.3 | 0 | — | — | — | — | |
| BeConciseBackbone=Qwen3-14B2025.12 | 76.4 | -54.2 | — | — | — | — | |
| AbbreWordsBackbone=Qwen3-14B2025.12 | 75.9 | -50.8 | — | — | — | — | |
| Qwen2.5-CoderModel Size=14B, GPU Hour=∼1.8M2025.05 | 75.08 | — | — | — | — | — | |
| NEMOTRON-CROSSTHINKCategory=Data Source, Blend=Bgpr↑2025.04 | 74.94 | — | — | — | — | — | |
| NEMOTRON-CROSSTHINKCategory=Data Source, Blend=Bmr↑2025.04 | 74.85 | — | — | — | — | — | |
| NEMOTRON-CROSSTHINKCategory=Data Usefulness, Blend=Bscore2025.04 | 74.7 | — | — | — | — | — | |
| NEMOTRON-CROSSTHINKCategory=Question Types, Blend=Bopen↑2025.04 | 74.46 | — | — | — | — | — | |
| NEMOTRON-CROSSTHINKCategory=Question Types, Blend=Bmcq↑2025.04 | 74.26 | — | — | — | — | — | |
| NEMOTRON-CROSSTHINKCategory=Single Source, Blend=Bonly_mr2025.04 | 74.24 | — | — | — | — | — | |
| OMACOptimization Dimension=Functional Dimension, Sub-dimension=Fun-1.4, Base LLM=gpt-3.5-turbo-1106, Temperature=0.82025.05 | 74.22 | — | — | — | — | — | |
| MModel=M2025.04 | 74.2 | — | — | — | — | — | |
| OMACOptimization Dimension=Functional Dimension, Sub-dimension=Fun-1.2, Base LLM=gpt-3.5-turbo-1106, Temperature=0.82025.05 | 73.23 | — | — | — | — | — | |
| ORZModel=ORZ2025.04 | 73.2 | — | — | — | — | — | |
| NEMOTRON-CROSSTHINKBlend=Bnd2025.04 | 73.18 | — | — | — | — | — | |
| OMACOptimization Dimension=Structural Dimension, Sub-dimension=Str-3, Base LLM=gpt-3.5-turbo-1106, Temperature=0.82025.05 | 73.18 | — | — | — | — | — | |
| OMACOptimization Dimension=Functional Dimension, Sub-dimension=Fun-1.5, Base LLM=gpt-3.5-turbo-1106, Temperature=0.82025.05 | 73.15 | — | — | — | — | — | |
| OMACOptimization Dimension=Structural Dimension, Sub-dimension=Str-1, Base LLM=gpt-3.5-turbo-1106, Temperature=0.82025.05 | 73.14 | — | — | — | — | — | |
| NEMOTRON-CROSSTHINKCategory=Single Source, Blend=Bonly_gpr2025.04 | 72.77 | — | — | — | — | — | |
| OMACOptimization Dimension=Functional Dimension, Sub-dimension=Fun-1.1, Base LLM=gpt-3.5-turbo-1106, Temperature=0.82025.05 | 72.33 | — | — | — | — | — | |
| OMACOptimization Dimension=Functional Dimension, Sub-dimension=Fun-1.6, Base LLM=gpt-3.5-turbo-1106, Temperature=0.82025.05 | 72.07 | — | — | — | — | — | |
| OMACOptimization Dimension=Structural Dimension, Sub-dimension=Str-2, Base LLM=gpt-3.5-turbo-1106, Temperature=0.82025.05 | 72.06 | — | — | — | — | — | |
| OMACOptimization Dimension=Functional Dimension, Sub-dimension=Fun-1.3, Base LLM=gpt-3.5-turbo-1106, Temperature=0.82025.05 | 72.06 | — | — | — | — | — | |
| OMACOptimization Dimension=Functional Dimension, Sub-dimension=Fun-1.7, Base LLM=gpt-3.5-turbo-1106, Temperature=0.82025.05 | 71.83 | — | — | — | — | — | |
| OMACOptimization Dimension=Functional Dimension, Sub-dimension=Fun-2, Base LLM=gpt-3.5-turbo-1106, Temperature=0.82025.05 | 71.02 | — | — | — | — | — | |
| Qwen2.5-7B2026.02 | 70.41 | — | — | — | — | — | |
| AFlowBase LLM=gpt-3.5-turbo-1106, Temperature=0.82025.05 | 70.06 | — | — | — | — | — | |
| DyLANBase LLM=gpt-3.5-turbo-1106, Temperature=0.82025.05 | 69.42 | — | — | — | — | — | |
| ADASBase LLM=gpt-3.5-turbo-1106, Temperature=0.82025.05 | 69.02 | — | — | — | — | — | |
| LLM DebateBase LLM=gpt-3.5-turbo-1106, Temperature=0.82025.05 | 68.74 | — | — | — | — | — | |
| SEBase LLM=gpt-3.5-turbo-1106, Temperature=0.82025.05 | 65.76 | — | — | — | — | — | |
| BioBridge2026.02 | 63.3 | — | — | — | — | — | |
| Prot2Chat2026.02 | 24.48 | — | — | — | — | — | |
| ProtT32026.02 | 4.29 | — | — | — | — | — | |
| EXPECTED-DIFFICULTYModel=Qwen3-4B2026.03 | — | — | 59.6 | 71.22 | 77.93 | 68.63 | |
| INSIGHTModel=Qwen3-0.6B2026.03 | — | — | 39.21 | 47.34 | 52.84 | 41.8 | |
| INSIGHTModel=Qwen3-1.7B2026.03 | — | — | 49 | 60.83 | 65 | 54.61 | |
| INSIGHTModel=Qwen3-4B2026.03 | — | — | 59.7 | 71.13 | 78.1 | 69.14 | |
| INVERSE-EVIDENCEModel=Qwen3-0.6B2026.03 | — | — | 38.55 | 45.19 | 49.76 | 36.95 | |
| INVERSE-EVIDENCEModel=Qwen3-1.7B2026.03 | — | — | 48.8 | 60.4 | 64.41 | 54.14 | |
| INVERSE-EVIDENCEModel=Qwen3-4B2026.03 | — | — | 59.36 | 71.12 | 78.03 | 68.79 | |
| MOPPSModel=Qwen3-0.6B2026.03 | — | — | 39.17 | 47.12 | 51.12 | 38.69 | |
| MOPPSModel=Qwen3-1.7B2026.03 | — | — | 48.63 | 60.32 | 64.38 | 53.35 | |
| MOPPSModel=Qwen3-4B2026.03 | — | — | 59.62 | 71 | 78 | 69.01 | |
| RANDOMModel=Qwen3-0.6B2026.03 | — | — | 38.83 | 47.22 | 52.32 | 40.95 | |
| RANDOMModel=Qwen3-1.7B2026.03 | — | — | 48.63 | 60.41 | 64.09 | 53.47 | |
| RANDOMModel=Qwen3-4B2026.03 | — | — | 59.13 | 71.07 | 77.87 | 68.25 |