Model comparison
Lilith-31B-v1.0 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 18,687,061,792. At long context the gap widens: Qwen3-30B-A3B-Thinking-2507's KV cache at 32K is 2.1× smaller, which usually matters more than the difference in weights.
From the file· summed bytes, KV per layer
Side by side
| Lilith-31B-v1.0 | Qwen3-30B-A3B-Thinking-2507 | |
|---|---|---|
| Parameters | 31.3B | 30.5B |
| Architecture | gemma4 | qwen3moe |
| Layers | 60 | 48 |
| Native context | 262,144 | 262,144 |
| Mixture of experts | no | yes, 128 experts |
| Quantizations published | 41 | 51 |
| Smallest quantization | 6.67 GiB | 7.05 GiB |
| Q4_K_M | 17.40 GiB | 17.28 GiB |
| Licence | gemma | apache-2.0 |
KV cache by context
the term that decides long-context viability
| Context | Lilith-31B-v1.0 | Qwen3-30B-A3B-Thinking-2507 | Ratio |
|---|---|---|---|
| 4,096 | 1.80 GiB | 0.38 GiB | 4.79× |
| 8,192 | 2.42 GiB | 0.75 GiB | 3.23× |
| 16,384 | 3.67 GiB | 1.50 GiB | 2.45× |
| 32,768 | 6.17 GiB | 3.00 GiB | 2.06× |
| 65,536 | 11.17 GiB | 6.00 GiB | 1.86× |
| 131,072 | 21.17 GiB | 12.00 GiB | 1.76× |