General Capability Evaluation on MMLU, GSM8K, HumanEval, IFEval
77.78Common Average ScoreNOVA
Evaluation results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| NOVAModel=Qwen3, Size=8B2026.04 | 77.78 | 76.43 | 90.75 | 82.93 | 61 | |
| PRISMModel=Qwen3, Size=8B2026.04 | 77.78 | 77.54 | 85.67 | 85.98 | 61.92 | |
| SEALModel=Qwen3, Size=8B2026.04 | 77.19 | 77.79 | 88.86 | 81.1 | 61 | |
| PRISMModel=Qwen3, Size=4B2026.04 | 77.06 | 76.93 | 86.2 | 81.71 | 63.4 | |
| SEALModel=Qwen3, Size=4B2026.04 | 76.09 | 76.01 | 87.64 | 79.88 | 60.81 | |
| SFTModel=Qwen3, Size=4B2026.04 | 75.35 | 76.19 | 85.14 | 79.27 | 60.81 | |
| NOVAModel=Qwen3, Size=4B2026.04 | 74.11 | 75.28 | 81.58 | 81.71 | 57.86 | |
| SFTModel=Qwen3, Size=8B2026.04 | 73.86 | 78.21 | 88.02 | 83.54 | 45.66 | |
| AlphaTokenModel=Qwen-3.5-9B2026.06 | 72.59 | — | — | — | — | |
| ssTOKEN ICLR 2026Model=Qwen-3.5-9B2026.06 | 70.08 | — | — | — | — | |
| XTF ICLR 2026Model=Qwen-3.5-9B2026.06 | 69.84 | — | — | — | — | |
| Token Cleaning ICML 2025Model=Qwen-3.5-9B2026.06 | 69.32 | — | — | — | — | |
| Standard FTModel=Qwen-3.5-9B2026.06 | 68.93 | — | — | — | — | |
| STM NeurIPS 2025Model=Qwen-3.5-9B2026.06 | 68.92 | — | — | — | — | |
| LESS ICML 2024Model=Qwen-3.5-9B2026.06 | 68.7 | — | — | — | — | |
| LoRA ICLR 2022Model=Qwen-3.5-9B2026.06 | 67.51 | — | — | — | — | |
| Pre-trainedModel=Qwen-3.5-9B2026.06 | 64.58 | — | — | — | — | |
| AlphaTokenModel=Gemma-3-4B2026.06 | 56.21 | — | — | — | — | |
| ssTOKEN ICLR 2026Model=Gemma-3-4B2026.06 | 53.41 | — | — | — | — | |
| Token Cleaning ICML 2025Model=Gemma-3-4B2026.06 | 53.02 | — | — | — | — | |
| LESS ICML 2024Model=Gemma-3-4B2026.06 | 52.77 | — | — | — | — | |
| XTF ICLR 2026Model=Gemma-3-4B2026.06 | 52.49 | — | — | — | — | |
| STM NeurIPS 2025Model=Gemma-3-4B2026.06 | 52.01 | — | — | — | — | |
| Standard FTModel=Gemma-3-4B2026.06 | 52 | — | — | — | — | |
| LoRA ICLR 2022Model=Gemma-3-4B2026.06 | 51.46 | — | — | — | — | |
| NOVAModel=Llama 3.1, Size=8B2026.04 | 49.14 | 49.96 | 58.76 | 51.22 | 36.6 | |
| PRISMModel=Llama 3.1, Size=8B2026.04 | 47.97 | 47.72 | 53.3 | 47.8 | 43.07 | |
| SFTModel=Llama 3.1, Size=8B2026.04 | 47.31 | 50.7 | 51.86 | 40.85 | 45.84 | |
| SEALModel=Llama 3.1, Size=8B2026.04 | 46.99 | 48.62 | 49.73 | 44.51 | 45.1 | |
| AlphaTokenModel=Llama-3.2-3B2026.06 | 44.73 | — | — | — | — | |
| Pre-trainedModel=Gemma-3-4B2026.06 | 43.3 | — | — | — | — | |
| ssTOKEN ICLR 2026Model=Llama-3.2-3B2026.06 | 43.19 | — | — | — | — | |
| STM NeurIPS 2025Model=Llama-3.2-3B2026.06 | 43.04 | — | — | — | — | |
| LESS ICML 2024Model=Llama-3.2-3B2026.06 | 42.8 | — | — | — | — | |
| Standard FTModel=Llama-3.2-3B2026.06 | 42.46 | — | — | — | — | |
| Token Cleaning ICML 2025Model=Llama-3.2-3B2026.06 | 42.21 | — | — | — | — | |
| XTF ICLR 2026Model=Llama-3.2-3B2026.06 | 41.73 | — | — | — | — | |
| LoRA ICLR 2022Model=Llama-3.2-3B2026.06 | 41.53 | — | — | — | — | |
| Pre-trainedModel=Llama-3.2-3B2026.06 | 37.18 | — | — | — | — |