Model comparison
Bonsai-8B-unpacked vs Qwen3-30B-A3B-Thinking-2507
At Q4_K_M, Bonsai-8B-unpacked is the smaller download — 5,198,494,208 bytes against 18,556,685,824. At long context the gap widens: Qwen3-30B-A3B-Thinking-2507's KV cache at 32K is 1.5× smaller, which usually matters more than the difference in weights.
From the file· summed bytes, KV per layer
Side by side
| Bonsai-8B-unpacked | Qwen3-30B-A3B-Thinking-2507 | |
|---|---|---|
| Parameters | 8.2B | 30.5B |
| Architecture | qwen3 | qwen3moe |
| Layers | 36 | 48 |
| Native context | 65,536 | 262,144 |
| Mixture of experts | no | yes, 128 experts |
| Quantizations published | 23 | 51 |
| Smallest quantization | 3.13 GiB | 7.05 GiB |
| Q4_K_M | 4.84 GiB | 17.28 GiB |
| Licence | apache-2.0 | apache-2.0 |
KV cache by context
the term that decides long-context viability
| Context | Bonsai-8B-unpacked | Qwen3-30B-A3B-Thinking-2507 | Ratio |
|---|---|---|---|
| 4,096 | 0.56 GiB | 0.38 GiB | 1.50× |
| 8,192 | 1.13 GiB | 0.75 GiB | 1.50× |
| 16,384 | 2.25 GiB | 1.50 GiB | 1.50× |
| 32,768 | 4.50 GiB | 3.00 GiB | 1.50× |
| 65,536 | 9.00 GiB | 6.00 GiB | 1.50× |
| 131,072 | 18.00 GiB | 12.00 GiB | 1.50× |