Model comparison

Qwen3-Coder-30B-A3B-Instruct vs Qwythos-9B-Claude-Mythos-5-1M-uncensored-heretic

At Q4_K_M, Qwythos-9B-Claude-Mythos-5-1M-uncensored-heretic is the smaller download — 6,216,967,936 bytes against 18,556,689,568. At long context the gap widens: Qwythos-9B-Claude-Mythos-5-1M-uncensored-heretic's KV cache at 32K is 3.0× smaller, which usually matters more than the difference in weights.

From the file· summed bytes, KV per layer

Side by side

Qwen3-Coder-30B-A3B-InstructQwythos-9B-Claude-Mythos-5-1M-uncensored-heretic
Parameters30.5B9.4B
Architectureqwen3moeqwen35
Layers4832
Native context262,1441,048,576
Mixture of expertsyes, 128 expertsno
Quantizations published4621
Smallest quantization7.46 GiB5.53 GiB
Q4_K_M17.28 GiB5.79 GiB
Licenceapache-2.0apache-2.0

KV cache by context

the term that decides long-context viability
ContextQwen3-Coder-30B-A3B-InstructQwythos-9B-Claude-Mythos-5-1M-uncensored-hereticRatio
4,0960.38 GiB0.13 GiB3.00×
8,1920.75 GiB0.25 GiB3.00×
16,3841.50 GiB0.50 GiB3.00×
32,7683.00 GiB1.00 GiB3.00×
65,5366.00 GiB2.00 GiB3.00×
131,07212.00 GiB4.00 GiB3.00×