Model comparison
Qwen3-16B-A3B vs gemma-4-E4B-it
At Q4_K_M, gemma-4-E4B-it is the smaller download — 4,977,171,584 bytes against 9,754,940,096. At long context the gap widens: gemma-4-E4B-it's KV cache at 32K is 5.9× smaller, which usually matters more than the difference in weights.
From the file· summed bytes, KV per layer
Side by side
| Qwen3-16B-A3B | gemma-4-E4B-it | |
|---|---|---|
| Parameters | 16.0B | 8.0B |
| Architecture | qwen3moe | gemma4 |
| Layers | 48 | 42 |
| Native context | 40,960 | 131,072 |
| Mixture of experts | yes, 64 experts | no |
| Quantizations published | 46 | 36 |
| Smallest quantization | 4.12 GiB | 3.30 GiB |
| Q4_K_M | 9.08 GiB | 4.64 GiB |
| Licence | apache-2.0 | apache-2.0 |
KV cache by context
the term that decides long-context viability
| Context | Qwen3-16B-A3B | gemma-4-E4B-it | Ratio |
|---|---|---|---|
| 4,096 | 0.38 GiB | 0.12 GiB | 3.05× |
| 8,192 | 0.75 GiB | 0.18 GiB | 4.22× |
| 16,384 | 1.50 GiB | 0.29 GiB | 5.22× |
| 32,768 | 3.00 GiB | 0.51 GiB | 5.93× |
| 65,536 | 6.00 GiB | 0.94 GiB | 6.36× |
| 131,072 | 12.00 GiB | 1.82 GiB | 6.60× |