Model comparison
gemma-4-12B-it vs gemma-3n-E4B-it
At Q4_K_M, gemma-3n-E4B-it is the smaller download — 4,237,064,064 bytes against 7,121,861,440. At long context the gap widens: gemma-3n-E4B-it's KV cache at 32K is 5.0× smaller, which usually matters more than the difference in weights.
From the file· summed bytes, KV per layer
Side by side
| gemma-4-12B-it | gemma-3n-E4B-it | |
|---|---|---|
| Parameters | 12.0B | 7.8B |
| Architecture | gemma4 | gemma3n |
| Layers | 48 | 35 |
| Native context | 262,144 | 32,768 |
| Mixture of experts | no | no |
| Quantizations published | 59 | 38 |
| Smallest quantization | 3.92 GiB | 2.57 GiB |
| Q4_K_M | 6.63 GiB | 3.95 GiB |
| Licence | apache-2.0 | gemma |
KV cache by context
the term that decides long-context viability
| Context | gemma-4-12B-it | gemma-3n-E4B-it | Ratio |
|---|---|---|---|
| 4,096 | 0.72 GiB | 0.11 GiB | 6.57× |
| 8,192 | 0.97 GiB | 0.16 GiB | 5.90× |
| 16,384 | 1.47 GiB | 0.27 GiB | 5.37× |
| 32,768 | 2.47 GiB | 0.49 GiB | 5.02× |
| 65,536 | 4.47 GiB | 0.93 GiB | 4.81× |
| 131,072 | 8.47 GiB | 1.80 GiB | 4.69× |