### Qwen3-8B (baseline: same-quant solo)

| Config | n | Math tok/s | Math x | Code tok/s | Code x | Chat tok/s | Chat x | All tok/s | All x | alpha all |
| -------- | --- | ------------ | -------- | ------------ | -------- | ------------ | -------- | ----------- | ------- | ----------- |
| q4-solo | 1474 | 53.5 | — | 53.2 | — | 53.4 | — | 53.4 | — | — |
| q4-vanilla17b | 1474 | 80.0 | 1.50x | 74.4 | 1.40x | 72.6 | 1.36x | 75.5 | 1.41x | 0.725 |
| q4-eagle3 | 1474 | 75.0 | 1.40x | 74.9 | 1.41x | 73.2 | 1.37x | 74.4 | 1.39x | 0.440 |
| q4-dflash | 1474 | 33.4 | 0.63x | 33.4 | 0.63x | 33.5 | 0.63x | 33.4 | 0.63x | 0.009 |
| q4-dspark-p0 | 1474 | 95.9 | 1.79x | 86.5 | 1.63x | 81.5 | 1.53x | 87.6 | 1.64x | 0.616 |
| q4-dspark-p2 | 1474 | 95.0 | 1.78x | 85.6 | 1.61x | 80.5 | 1.51x | 86.7 | 1.62x | 0.621 |
| q4-dspark-p4 | 1474 | 94.6 | 1.77x | 84.8 | 1.59x | 77.9 | 1.46x | 85.4 | 1.60x | 0.646 |
| q4-dspark-p6 | 1474 | 92.7 | 1.73x | 79.7 | 1.50x | 71.3 | 1.33x | 80.7 | 1.51x | 0.714 |
| q5-solo | 1474 | 46.9 | — | 46.7 | — | 46.8 | — | 46.8 | — | — |
| q5-eagle3 | 1474 | 69.0 | 1.47x | 69.1 | 1.48x | 66.7 | 1.43x | 68.3 | 1.46x | 0.439 |
| q5-dflash | 1474 | 30.8 | 0.66x | 30.7 | 0.66x | 30.8 | 0.66x | 30.8 | 0.66x | 0.009 |
| q5-dspark-p0 | 1474 | 87.9 | 1.87x | 80.1 | 1.72x | 74.7 | 1.60x | 80.6 | 1.72x | 0.613 |
| q8-solo | 1474 | 32.4 | — | 32.3 | — | 32.3 | — | 32.3 | — | — |
| q8-eagle3 | 1474 | 53.5 | 1.65x | 53.4 | 1.65x | 51.5 | 1.59x | 52.8 | 1.63x | 0.432 |
| q8-dspark-p0 | 1474 | 68.4 | 2.11x | 61.8 | 1.91x | 57.5 | 1.78x | 62.3 | 1.93x | 0.611 |

### Gemma 4 12B (baseline: same-quant solo)

| Config | n | Math tok/s | Math x | Code tok/s | Code x | Chat tok/s | Chat x | All tok/s | All x | alpha all |
| -------- | --- | ------------ | -------- | ------------ | -------- | ------------ | -------- | ----------- | ------- | ----------- |
| q4-solo | 1455 | 34.7 | — | 34.4 | — | 34.9 | — | 34.7 | — | — |
| q4-dflash-f16 | 1455 | 80.8 | 2.33x | 76.3 | 2.22x | 59.4 | 1.71x | 72.2 | 2.09x | 0.657 |
| q4-dflash-q4 | 1455 | 88.1 | 2.54x | 82.8 | 2.41x | 64.2 | 1.85x | 78.4 | 2.26x | 0.647 |
| q4-dflash-q8 | 1455 | 85.7 | 2.47x | 81.0 | 2.35x | 63.0 | 1.81x | 76.6 | 2.21x | 0.656 |
| q4-mtp | 1455 | 89.8 | 2.59x | 85.6 | 2.49x | 72.6 | 2.09x | 82.7 | 2.39x | 0.739 |
| q5-solo | 1449 | 29.9 | — | 29.7 | — | 30.0 | — | 29.9 | — | — |
| q5-dflash-f16 | 1449 | 72.5 | 2.42x | 68.4 | 2.30x | 52.5 | 1.75x | 64.5 | 2.16x | 0.659 |
| q5-mtp | 1449 | 82.3 | 2.75x | 78.4 | 2.64x | 66.3 | 2.21x | 75.6 | 2.53x | 0.745 |
| q8-solo | 1452 | 19.9 | — | 19.8 | — | 20.1 | — | 19.9 | — | — |
| q8-dflash-f16 | 1452 | 51.3 | 2.58x | 48.7 | 2.46x | 36.9 | 1.85x | 45.7 | 2.30x | 0.658 |
| q8-mtp | 1452 | 58.5 | 2.94x | 56.0 | 2.82x | 47.2 | 2.36x | 54.0 | 2.71x | 0.745 |