Model comparison
Qwen3-0.6B-heretic-abliterated-uncensored vs MiniCPM5-1B-Claude-Opus-Fable5-Thinking
At Q4_K_M, Qwen3-0.6B-heretic-abliterated-uncensored is the smaller download — 396,705,312 bytes against 688,066,496. At long context the gap widens: MiniCPM5-1B-Claude-Opus-Fable5-Thinking's KV cache at 32K is 4.7× smaller, which usually matters more than the difference in weights.
From the file· summed bytes, KV per layer
Side by side
| Qwen3-0.6B-heretic-abliterated-uncensored | MiniCPM5-1B-Claude-Opus-Fable5-Thinking | |
|---|---|---|
| Parameters | 596M | 1.1B |
| Architecture | qwen3 | llama |
| Layers | 28 | 24 |
| Native context | 40,960 | 131,072 |
| Mixture of experts | no | no |
| Quantizations published | 36 | 15 |
| Smallest quantization | 0.19 GiB | 0.43 GiB |
| Q4_K_M | 0.37 GiB | 0.64 GiB |
| Licence | — | apache-2.0 |
KV cache by context
the term that decides long-context viability
| Context | Qwen3-0.6B-heretic-abliterated-uncensored | MiniCPM5-1B-Claude-Opus-Fable5-Thinking | Ratio |
|---|---|---|---|
| 4,096 | 0.44 GiB | 0.09 GiB | 4.67× |
| 8,192 | 0.88 GiB | 0.19 GiB | 4.67× |
| 16,384 | 1.75 GiB | 0.38 GiB | 4.67× |
| 32,768 | 3.50 GiB | 0.75 GiB | 4.67× |
| 65,536 | 7.00 GiB | 1.50 GiB | 4.67× |
| 131,072 | 14.00 GiB | 3.00 GiB | 4.67× |