Model comparison
NVIDIA-Nemotron-3-Nano-30B-A3B-BF16 vs Qwen3-30B-A3B-Thinking-2507
At Q4_K_M, Qwen3-30B-A3B-Thinking-2507 is the smaller download — 18,556,685,824 bytes against 22,421,827,488. At long context the gap widens: NVIDIA-Nemotron-3-Nano-30B-A3B-BF16's KV cache at 32K is 1.8× smaller, which usually matters more than the difference in weights.
From the file· summed bytes, KV per layer
Side by side
| NVIDIA-Nemotron-3-Nano-30B-A3B-BF16 | Qwen3-30B-A3B-Thinking-2507 | |
|---|---|---|
| Parameters | 31.6B | 30.5B |
| Architecture | nemotron_h_moe | qwen3moe |
| Layers | 52 | 48 |
| Native context | 262,144 | 262,144 |
| Mixture of experts | yes, 128 experts | yes, 128 experts |
| Quantizations published | 25 | 51 |
| Smallest quantization | 16.85 GiB | 7.05 GiB |
| Q4_K_M | 20.88 GiB | 17.28 GiB |
| Licence | other | apache-2.0 |
KV cache by context
the term that decides long-context viability
| Context | NVIDIA-Nemotron-3-Nano-30B-A3B-BF16 | Qwen3-30B-A3B-Thinking-2507 | Ratio |
|---|---|---|---|
| 4,096 | 0.20 GiB | 0.38 GiB | 1.85× |
| 8,192 | 0.41 GiB | 0.75 GiB | 1.85× |
| 16,384 | 0.81 GiB | 1.50 GiB | 1.85× |
| 32,768 | 1.63 GiB | 3.00 GiB | 1.85× |
| 65,536 | 3.25 GiB | 6.00 GiB | 1.85× |
| 131,072 | 6.50 GiB | 12.00 GiB | 1.85× |