Model comparison

Qwen3-Coder-30B-A3B-Instruct vs Qwen3.5-40B-RoughHouse-Claude-4.6-Opus-Polar-Deckard-Uncensored-Heretic-Thinking

At Q4_K_M, Qwen3-Coder-30B-A3B-Instruct is the smaller download — 18,556,689,568 bytes against 23,936,565,984.

From the file· summed bytes, KV per layer

Side by side

Qwen3-Coder-30B-A3B-InstructQwen3.5-40B-RoughHouse-Claude-4.6-Opus-Polar-Deckard-Uncensored-Heretic-Thinking
Parameters30.5B39.5B
Architectureqwen3moeqwen35
Layers4896
Native context262,144262,144
Mixture of expertsyes, 128 expertsno
Quantizations published4634
Smallest quantization7.46 GiB9.38 GiB
Q4_K_M17.28 GiB22.29 GiB
Licenceapache-2.0apache-2.0

KV cache by context

the term that decides long-context viability
ContextQwen3-Coder-30B-A3B-InstructQwen3.5-40B-RoughHouse-Claude-4.6-Opus-Polar-Deckard-Uncensored-Heretic-ThinkingRatio
4,0960.38 GiB0.38 GiB1.00×
8,1920.75 GiB0.75 GiB1.00×
16,3841.50 GiB1.50 GiB1.00×
32,7683.00 GiB3.00 GiB1.00×
65,5366.00 GiB6.00 GiB1.00×
131,07212.00 GiB12.00 GiB1.00×
Qwen3-Coder-30B-A3B-Instruct vs Qwen3.5-40B-RoughHouse-Claude-4.6-Opus-Polar-Deckard-Uncensored-Heretic-Thinking — size, memory and hardware fit — ossmodeldb