Original is shown as a reference and is excluded from bold best-score marking. Bold values mark the best quantized score within each model for each benchmark, including the average; ties are all bolded.| Model | Quantization | MMLU 1000 | Winogrande 300 | HumanEval 164 | MBPP 300 | Average % |
|---|---|---|---|---|---|---|
| Qwen3.6-27B | Original (55.6GB) | 877/1000 87.70% | 241/300 80.33% | 149/164 90.85% | 250/300 83.33% | 85.56% |
| mlx-lm 4bit (16.1GB) | 865/1000 86.50% | 239/300 79.67% | 153/164 93.29% | 244/300 81.33% | 85.20% | |
| oQ4 (16.7GB) | 869/1000 86.90% | 239/300 79.67% | 151/164 92.07% | 244/300 81.33% | 84.99% | |
| oQ4e imatrix (16.7GB) | 868/1000 86.80% | 239/300 79.67% | 152/164 92.68% | 250/300 83.33% | 85.62% |