Model comparison

Qwen3-Coder-30B-A3B-Instruct vs Octen-Embedding-8B

At Q4_K_M, Octen-Embedding-8B is the smaller download — 5,409,650,336 bytes against 18,556,689,568. At long context the gap widens: Qwen3-Coder-30B-A3B-Instruct's KV cache at 32K is 1.5× 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-InstructOcten-Embedding-8B
Parameters30.5B7.6B
Architectureqwen3moeqwen3
Layers4836
Native context262,14440,960
Mixture of expertsyes, 128 expertsno
Quantizations published467
Smallest quantization7.46 GiB4.06 GiB
Q4_K_M17.28 GiB5.04 GiB
Licenceapache-2.0apache-2.0

KV cache by context

the term that decides long-context viability
ContextQwen3-Coder-30B-A3B-InstructOcten-Embedding-8BRatio
4,0960.38 GiB0.56 GiB1.50×
8,1920.75 GiB1.13 GiB1.50×
16,3841.50 GiB2.25 GiB1.50×
32,7683.00 GiB4.50 GiB1.50×
65,5366.00 GiB9.00 GiB1.50×
131,07212.00 GiB18.00 GiB1.50×
Qwen3-Coder-30B-A3B-Instruct vs Octen-Embedding-8B — size, memory and hardware fit — ossmodeldb