Model comparison
NVIDIA-Nemotron-Nano-9B-v2 vs Qwen3.8-27B
At Q4_K_M, NVIDIA-Nemotron-Nano-9B-v2 is the smaller download — 6,525,628,992 bytes against 16,810,714,336. At long context the gap widens: Qwen3.8-27B's KV cache at 32K is 3.5× smaller, which usually matters more than the difference in weights.
From the file· summed bytes, KV per layer
Side by side
| NVIDIA-Nemotron-Nano-9B-v2 | Qwen3.8-27B | |
|---|---|---|
| Parameters | 8.9B | 27.8B |
| Architecture | nemotron_h | qwen35 |
| Layers | 56 | 64 |
| Native context | 131,072 | 262,144 |
| Mixture of experts | no | no |
| Quantizations published | 34 | 22 |
| Smallest quantization | 4.62 GiB | 8.39 GiB |
| Q4_K_M | 6.08 GiB | 15.66 GiB |
| Licence | — | apache-2.0 |
KV cache by context
the term that decides long-context viability
| Context | NVIDIA-Nemotron-Nano-9B-v2 | Qwen3.8-27B | Ratio |
|---|---|---|---|
| 4,096 | 0.88 GiB | 0.25 GiB | 3.50× |
| 8,192 | 1.75 GiB | 0.50 GiB | 3.50× |
| 16,384 | 3.50 GiB | 1.00 GiB | 3.50× |
| 32,768 | 7.00 GiB | 2.00 GiB | 3.50× |
| 65,536 | 14.00 GiB | 4.00 GiB | 3.50× |
| 131,072 | 28.00 GiB | 8.00 GiB | 3.50× |