Model comparison
Kimi-K2.7-Code vs Inkling
These two publish different quantization sets; the table below has the exact sizes. At long context the gap widens: Kimi-K2.7-Code's KV cache at 32K is 3.8× smaller, which usually matters more than the difference in weights.
From the file· summed bytes, KV per layer
Side by side
| Kimi-K2.7-Code | Inkling | |
|---|---|---|
| Parameters | 1059B | 952B |
| Architecture | deepseek2 | inkling |
| Layers | 61 | 66 |
| Native context | 262,144 | — |
| Mixture of experts | yes, 384 experts | yes, 256 experts |
| Quantizations published | 19 | 7 |
| Smallest quantization | 283.04 GiB | 210.69 GiB |
| Q4_K_M | — | — |
| Licence | other | apache-2.0 |
KV cache by context
the term that decides long-context viability
| Context | Kimi-K2.7-Code | Inkling | Ratio |
|---|---|---|---|
| 4,096 | 0.27 GiB | 1.03 GiB | 3.85× |
| 8,192 | 0.54 GiB | 2.06 GiB | 3.85× |
| 16,384 | 1.07 GiB | 4.13 GiB | 3.85× |
| 32,768 | 2.14 GiB | 8.25 GiB | 3.85× |
| 65,536 | 4.29 GiB | 16.50 GiB | 3.85× |
| 131,072 | 8.58 GiB | 33.00 GiB | 3.85× |