Model comparison
diffusiongemma-26B-A4B-it vs Qwen3.6-35B-A3B
At Q4_K_M, diffusiongemma-26B-A4B-it is the smaller download — 16,806,810,208 bytes against 21,166,757,728. At long context the gap widens: Qwen3.6-35B-A3B's KV cache at 32K is 2.5× smaller, which usually matters more than the difference in weights.
From the file· summed bytes, KV per layer
Side by side
| diffusiongemma-26B-A4B-it | Qwen3.6-35B-A3B | |
|---|---|---|
| Parameters | 25.8B | 36.0B |
| Architecture | diffusion-gemma | qwen35moe |
| Layers | 30 | 40 |
| Native context | 262,144 | 262,144 |
| Mixture of experts | yes, 128 experts | yes, 256 experts |
| Quantizations published | 12 | 61 |
| Smallest quantization | 9.86 GiB | 8.77 GiB |
| Q4_K_M | 15.65 GiB | 19.71 GiB |
| Licence | apache-2.0 | apache-2.0 |
KV cache by context
the term that decides long-context viability
| Context | diffusiongemma-26B-A4B-it | Qwen3.6-35B-A3B | Ratio |
|---|---|---|---|
| 4,096 | 0.45 GiB | 0.08 GiB | 5.75× |
| 8,192 | 0.61 GiB | 0.16 GiB | 3.87× |
| 16,384 | 0.92 GiB | 0.31 GiB | 2.94× |
| 32,768 | 1.54 GiB | 0.63 GiB | 2.47× |
| 65,536 | 2.79 GiB | 1.25 GiB | 2.23× |
| 131,072 | 5.29 GiB | 2.50 GiB | 2.12× |