Model comparison

Llama-3_3-Nemotron-Super-49B-v1 vs Qwen3-30B-A3B-Thinking-2507

At Q4_K_M, Qwen3-30B-A3B-Thinking-2507 is the smaller download — 18,556,685,824 bytes against 30,215,576,224. At long context the gap widens: Qwen3-30B-A3B-Thinking-2507's KV cache at 32K is 26.7× smaller, which usually matters more than the difference in weights.

From the file· summed bytes, KV per layer

Side by side

Llama-3_3-Nemotron-Super-49B-v1Qwen3-30B-A3B-Thinking-2507
Parameters49.9B30.5B
Architecturedeciqwen3moe
Layers8048
Native context131,072262,144
Mixture of expertsnoyes, 128 experts
Quantizations published6651
Smallest quantization10.27 GiB7.05 GiB
Q4_K_M28.14 GiB17.28 GiB
Licenceotherapache-2.0

KV cache by context

the term that decides long-context viability
ContextLlama-3_3-Nemotron-Super-49B-v1Qwen3-30B-A3B-Thinking-2507Ratio
4,09610.00 GiB0.38 GiB26.67×
8,19220.00 GiB0.75 GiB26.67×
16,38440.00 GiB1.50 GiB26.67×
32,76880.00 GiB3.00 GiB26.67×
65,536160.00 GiB6.00 GiB26.67×
131,072320.00 GiB12.00 GiB26.67×
Llama-3_3-Nemotron-Super-49B-v1 vs Qwen3-30B-A3B-Thinking-2507 — size, memory and hardware fit — ossmodeldb