Model comparison

Qwen3-8B vs Llama-3.1-8B-Instruct

At Q4_K_M, Llama-3.1-8B-Instruct is the smaller download — 4,920,739,168 bytes against 5,027,783,488.

From the file· summed bytes, KV per layer

Side by side

Qwen3-8BLlama-3.1-8B-Instruct
Parameters8.2B8.0B
Architectureqwen3llama
Layers3632
Native context40,960131,072
Mixture of expertsnono
Quantizations published5145
Smallest quantization2.12 GiB2.02 GiB
Q4_K_M4.68 GiB4.58 GiB
Licenceapache-2.0llama3.1

KV cache by context

the term that decides long-context viability
ContextQwen3-8BLlama-3.1-8B-InstructRatio
4,0960.56 GiB0.50 GiB1.13×
8,1921.13 GiB1.00 GiB1.13×
16,3842.25 GiB2.00 GiB1.13×
32,7684.50 GiB4.00 GiB1.13×
65,5369.00 GiB8.00 GiB1.13×
131,07218.00 GiB16.00 GiB1.13×