Model comparison
NVIDIA-Nemotron-Nano-12B-v2 vs Qwen3-30B-A3B-Thinking-2507
At Q4_K_M, NVIDIA-Nemotron-Nano-12B-v2 is the smaller download — 7,494,497,504 bytes against 18,556,685,824. At long context the gap widens: Qwen3-30B-A3B-Thinking-2507's KV cache at 32K is 2.6× smaller, which usually matters more than the difference in weights.
From the file· summed bytes, KV per layer
Side by side
| NVIDIA-Nemotron-Nano-12B-v2 | Qwen3-30B-A3B-Thinking-2507 | |
|---|---|---|
| Parameters | 12.3B | 30.5B |
| Architecture | nemotron_h | qwen3moe |
| Layers | 62 | 48 |
| Native context | 131,072 | 262,144 |
| Mixture of experts | no | yes, 128 experts |
| Quantizations published | 30 | 51 |
| Smallest quantization | 3.79 GiB | 7.05 GiB |
| Q4_K_M | 6.98 GiB | 17.28 GiB |
| Licence | — | apache-2.0 |
KV cache by context
the term that decides long-context viability
| Context | NVIDIA-Nemotron-Nano-12B-v2 | Qwen3-30B-A3B-Thinking-2507 | Ratio |
|---|---|---|---|
| 4,096 | 0.97 GiB | 0.38 GiB | 2.58× |
| 8,192 | 1.94 GiB | 0.75 GiB | 2.58× |
| 16,384 | 3.88 GiB | 1.50 GiB | 2.58× |
| 32,768 | 7.75 GiB | 3.00 GiB | 2.58× |
| 65,536 | 15.50 GiB | 6.00 GiB | 2.58× |
| 131,072 | 31.00 GiB | 12.00 GiB | 2.58× |