Language Modeling on LLM Evaluation Suite
83.2AccuracyDefault
Evaluation Results
| Method | Links | ||||||||
|---|---|---|---|---|---|---|---|---|---|
| DefaultModel=Qwen-72B2026.03 | 83.2 | 192.5 | 145.2 | 4.82 | 1 | — | — | — | |
| AdaptiveEfficientLLMModel=Qwen-72B2026.03 | 83 | 98.5 | 72.1 | 2.28 | 2.15 | — | — | — | |
| DefaultModel=LLaMA-2-70B2026.03 | 82.5 | 185.2 | 138.5 | 4.52 | 1 | — | — | — | |
| Best Single-StageModel=Qwen-72B2026.03 | 82.5 | 132.8 | 98.5 | 3.28 | 1.45 | — | — | — | |
| AdaptiveEfficientLLMModel=LLaMA-2-70B2026.03 | 82.3 | 92.5 | 68.2 | 2.15 | 2.12 | — | — | — | |
| EfficientLLM Rec.Model=Qwen-72B2026.03 | 82.2 | 128.5 | 95.5 | 3.12 | 1.52 | — | — | — | |
| Manual SelectionModel=Qwen-72B2026.03 | 82 | 125.2 | 92.8 | 3.05 | 1.58 | — | — | — | |
| Best Single-StageModel=LLaMA-2-70B2026.03 | 81.8 | 128.5 | 95.2 | 3.15 | 1.48 | — | — | — | |
| DefaultModel=Mixtral-8x7B2026.03 | 81.8 | 165.8 | 98.5 | 3.85 | 1 | — | — | — | |
| EfficientLLM Rec.Model=LLaMA-2-70B2026.03 | 81.5 | 122.8 | 91.2 | 2.98 | 1.55 | — | — | — | |
| AdaptiveEfficientLLMModel=Mixtral-8x7B2026.03 | 81.5 | 85.2 | 52.8 | 1.95 | 2.05 | — | — | — | |
| Manual SelectionModel=LLaMA-2-70B2026.03 | 81.2 | 118.5 | 88.5 | 2.85 | 1.62 | — | — | — | |
| Best Single-StageModel=Mixtral-8x7B2026.03 | 81.2 | 115.2 | 72.8 | 2.75 | 1.42 | — | — | — | |
| EfficientLLM Rec.Model=Mixtral-8x7B2026.03 | 81 | 112.5 | 70.2 | 2.65 | 1.48 | — | — | — | |
| Manual SelectionModel=Mixtral-8x7B2026.03 | 80.8 | 108.5 | 68.5 | 2.52 | 1.55 | — | — | — | |
| DefaultModel=LLaMA-3-8B2026.03 | 72.1 | 48.5 | 15.2 | 0.95 | 1 | — | — | — | |
| AdaptiveEfficientLLMModel=LLaMA-3-8B2026.03 | 71.8 | 27.2 | 8.5 | 0.45 | 1.98 | — | — | — | |
| Best Single-StageModel=LLaMA-3-8B2026.03 | 71.5 | 35.2 | 11.8 | 0.72 | 1.35 | — | — | — | |
| DefaultModel=Mistral-7B2026.03 | 71.2 | 42.8 | 14.1 | 0.88 | 1 | — | — | — | |
| EfficientLLM Rec.Model=LLaMA-3-8B2026.03 | 71.2 | 33.5 | 11 | 0.68 | 1.46 | — | — | — | |
| AdaptiveEfficientLLMModel=Mistral-7B2026.03 | 71 | 23.5 | 7.9 | 0.39 | 2.02 | — | — | — | |
| Manual SelectionModel=LLaMA-3-8B2026.03 | 71 | 32.8 | 10.5 | 0.65 | 1.52 | — | — | — | |
| Best Single-StageModel=Mistral-7B2026.03 | 70.8 | 30.5 | 10.8 | 0.65 | 1.38 | — | — | — | |
| EfficientLLM Rec.Model=Mistral-7B2026.03 | 70.5 | 29.1 | 10.2 | 0.61 | 1.48 | — | — | — | |
| Manual SelectionModel=Mistral-7B2026.03 | 70.2 | 28.2 | 9.8 | 0.58 | 1.55 | — | — | — | |
| DefaultModel=Average Across All Models2026.03 | 69.9 | 88.6 | 53.9 | 2.01 | 1 | — | — | — | |
| AdaptiveEfficientLLMModel=Average Across All Models2026.03 | 69.6 | 46.3 | 27.7 | 0.97 | 1.98 | — | — | — | |
| Best Single-StageModel=Average Across All Models2026.03 | 69.3 | 62.3 | 38.3 | 1.44 | 1.4 | — | — | — | |
| EfficientLLM Rec.Model=Average Across All Models2026.03 | 69 | 60 | 36.6 | 1.35 | 1.47 | — | — | — | |
| Manual SelectionModel=Average Across All Models2026.03 | 68.9 | 58.3 | 35.5 | 1.3 | 1.54 | — | — | — | |
| DefaultModel=LLaMA-2-7B2026.03 | 68.5 | 45.2 | 13.5 | 0.85 | 1 | — | — | — | |
| AdaptiveEfficientLLMModel=LLaMA-2-7B2026.03 | 68.2 | 25.8 | 8.1 | 0.42 | 1.95 | — | — | — | |
| Best Single-StageModel=LLaMA-2-7B2026.03 | 67.9 | 32.8 | 10.2 | 0.62 | 1.42 | — | — | — | |
| Manual SelectionModel=LLaMA-2-7B2026.03 | 67.3 | 30.5 | 9.5 | 0.55 | 1.58 | — | — | — | |
| EfficientLLM Rec.Model=LLaMA-2-7B2026.03 | 67.1 | 31.2 | 9.8 | 0.58 | 1.52 | — | — | — | |
| DefaultModel=Phi-22026.03 | 56.8 | 18.3 | 4.2 | 0.15 | 1 | — | — | — | |
| AdaptiveEfficientLLMModel=Phi-22026.03 | 56.5 | 11.5 | 2.8 | 0.08 | 1.82 | — | — | — | |
| Best Single-StageModel=Phi-22026.03 | 56.2 | 14.1 | 3.5 | 0.11 | 1.34 | — | — | — | |
| Manual SelectionModel=Phi-22026.03 | 55.9 | 13.2 | 3.2 | 0.1 | 1.48 | — | — | — | |
| EfficientLLM Rec.Model=Phi-22026.03 | 55.5 | 13.8 | 3.4 | 0.1 | 1.4 | — | — | — | |
| Dense TransformerModel scale=2B params, Training tokens=40B tokens, Sparse=No2026.03 | 49.1 | — | 46.7 | — | — | 87.8 | 7.85 | 22.4 | |
| Sparse TransformerModel scale=2B params, Training tokens=40B tokens, Sparse=Yes2026.03 | 48.8 | — | 57.1 | — | — | 106 | 6.51 | 27.3 | |
| Dense TransformerModel scale=1.5B params, Training tokens=30B tokens, Sparse=No2026.03 | 46.4 | — | 62.8 | — | — | 119 | 5.73 | 31.8 | |
| Sparse TransformerModel scale=1.5B params, Training tokens=30B tokens, Sparse=Yes2026.03 | 46.2 | — | 45.1 | — | — | 141 | 4.87 | 35.5 | |
| Sparse TransformerModel scale=1B params, Training tokens=20B tokens, Sparse=Yes2026.03 | 44.7 | — | 33.1 | — | — | 219 | 3.17 | 52.1 | |
| Dense TransformerModel scale=1B params, Training tokens=20B tokens, Sparse=No2026.03 | 44.6 | — | 44.5 | — | — | 185 | 3.71 | 48.6 | |
| DefaultModel=LLaMA-2-1B2026.03 | 43.2 | 12.5 | 2.1 | 0.08 | 1 | — | — | — | |
| AdaptiveEfficientLLMModel=LLaMA-2-1B2026.03 | 43 | 7.8 | 1.4 | 0.04 | 1.72 | — | — | — | |
| Best Single-StageModel=LLaMA-2-1B2026.03 | 42.8 | 9.2 | 1.8 | 0.06 | 1.31 | — | — | — | |
| Manual SelectionModel=LLaMA-2-1B2026.03 | 42.5 | 8.5 | 1.6 | 0.05 | 1.45 | — | — | — | |
| EfficientLLM Rec.Model=LLaMA-2-1B2026.03 | 42.1 | 8.8 | 1.7 | 0.05 | 1.38 | — | — | — | |
| Dense TransformerModel scale=0.5B params, Training tokens=10B tokens, Sparse=No2026.03 | 40.4 | — | 26.2 | — | — | 410 | 1.63 | 97.3 | |
| Sparse TransformerModel scale=0.5B params, Training tokens=10B tokens, Sparse=Yes2026.03 | 40.4 | — | 21.2 | — | — | 480 | 1.43 | 95.9 |