Model comparison
reka-flash-3.1 vs gemma-4-E4B-it
At Q4_K_M, gemma-4-E4B-it is the smaller download — 4,977,171,584 bytes against 13,610,362,432. At long context the gap widens: gemma-4-E4B-it's KV cache at 32K is 8.2× smaller, which usually matters more than the difference in weights.
From the file· summed bytes, KV per layer
Side by side
| reka-flash-3.1 | gemma-4-E4B-it | |
|---|---|---|
| Parameters | 20.9B | 8.0B |
| Architecture | llama | gemma4 |
| Layers | 44 | 42 |
| Native context | 98,304 | 131,072 |
| Mixture of experts | no | no |
| Quantizations published | 51 | 36 |
| Smallest quantization | 6.15 GiB | 3.30 GiB |
| Q4_K_M | 12.68 GiB | 4.64 GiB |
| Licence | apache-2.0 | apache-2.0 |
KV cache by context
the term that decides long-context viability
| Context | reka-flash-3.1 | gemma-4-E4B-it | Ratio |
|---|---|---|---|
| 4,096 | 0.52 GiB | 0.12 GiB | 4.19× |
| 8,192 | 1.03 GiB | 0.18 GiB | 5.80× |
| 16,384 | 2.06 GiB | 0.29 GiB | 7.18× |
| 32,768 | 4.13 GiB | 0.51 GiB | 8.15× |
| 65,536 | 8.25 GiB | 0.94 GiB | 8.75× |
| 131,072 | 16.50 GiB | 1.82 GiB | 9.07× |