Model comparison
Llama3.2-30B-A3B-II-Dark-Champion-INSTRUCT-Heretic-Abliterated-Uncensored vs Qwen3-30B-A3B-Thinking-2507
At Q4_K_M, Llama3.2-30B-A3B-II-Dark-Champion-INSTRUCT-Heretic-Abliterated-Uncensored is the smaller download — 18,366,980,192 bytes against 18,556,685,824. At long context the gap widens: Qwen3-30B-A3B-Thinking-2507's KV cache at 32K is 2.0× smaller, which usually matters more than the difference in weights.
From the file· summed bytes, KV per layer
Side by side
| Llama3.2-30B-A3B-II-Dark-Champion-INSTRUCT-Heretic-Abliterated-Uncensored | Qwen3-30B-A3B-Thinking-2507 | |
|---|---|---|
| Parameters | 30.0B | 30.5B |
| Architecture | llama | qwen3moe |
| Layers | 47 | 48 |
| Native context | 131,072 | 262,144 |
| Mixture of experts | yes, 8 experts | yes, 128 experts |
| Quantizations published | 34 | 51 |
| Smallest quantization | 6.07 GiB | 7.05 GiB |
| Q4_K_M | 17.11 GiB | 17.28 GiB |
| Licence | apache-2.0 | apache-2.0 |
KV cache by context
the term that decides long-context viability
| Context | Llama3.2-30B-A3B-II-Dark-Champion-INSTRUCT-Heretic-Abliterated-Uncensored | Qwen3-30B-A3B-Thinking-2507 | Ratio |
|---|---|---|---|
| 4,096 | 0.73 GiB | 0.38 GiB | 1.96× |
| 8,192 | 1.47 GiB | 0.75 GiB | 1.96× |
| 16,384 | 2.94 GiB | 1.50 GiB | 1.96× |
| 32,768 | 5.88 GiB | 3.00 GiB | 1.96× |
| 65,536 | 11.75 GiB | 6.00 GiB | 1.96× |
| 131,072 | 23.50 GiB | 12.00 GiB | 1.96× |