Model comparison
Qwen3-15B-A2B-Base vs gemma-4-E4B-it
At Q4_K_M, gemma-4-E4B-it is the smaller download — 4,977,171,584 bytes against 9,496,453,792. At long context the gap widens: gemma-4-E4B-it's KV cache at 32K is 3.0× smaller, which usually matters more than the difference in weights.
From the file· summed bytes, KV per layer
Side by side
| Qwen3-15B-A2B-Base | gemma-4-E4B-it | |
|---|---|---|
| Parameters | 15.6B | 8.0B |
| Architecture | qwen3moe | gemma4 |
| Layers | 24 | 42 |
| Native context | 4,096 | 131,072 |
| Mixture of experts | yes, 128 experts | no |
| Quantizations published | 11 | 36 |
| Smallest quantization | 5.41 GiB | 3.30 GiB |
| Q4_K_M | 8.84 GiB | 4.64 GiB |
| Licence | apache-2.0 | apache-2.0 |
KV cache by context
the term that decides long-context viability
| Context | Qwen3-15B-A2B-Base | gemma-4-E4B-it | Ratio |
|---|---|---|---|
| 4,096 | 0.19 GiB | 0.12 GiB | 1.52× |
| 8,192 | 0.38 GiB | 0.18 GiB | 2.11× |
| 16,384 | 0.75 GiB | 0.29 GiB | 2.61× |
| 32,768 | 1.50 GiB | 0.51 GiB | 2.97× |
| 65,536 | 3.00 GiB | 0.94 GiB | 3.18× |
| 131,072 | 6.00 GiB | 1.82 GiB | 3.30× |