Is ultravox-v0_5-llama-3_2-1b or Qwen3-4B smaller?

ultravox-v0_5-llama-3_2-1b is smaller at Q4_K_M: 0.75 GiB against 2.33 GiB.

Which uses less memory at long context?

Qwen3-4B — its KV cache at 32K context is 1.8x smaller, because of how its attention layers cache.

Model comparison

ultravox-v0_5-llama-3_2-1b vs Qwen3-4B

At Q4_K_M, ultravox-v0_5-llama-3_2-1b is the smaller download — 807,694,464 bytes against 2,497,280,256. At long context the gap widens: Qwen3-4B's KV cache at 32K is 1.8× smaller, which usually matters more than the difference in weights.

From the file· summed bytes, KV per layer

Side by side

	ultravox-v0_5-llama-3_2-1b	Qwen3-4B
Parameters	683M	4.0B
Architecture	llama	qwen3
Layers	16	36
Native context	—	40,960
Mixture of experts	no	no
Quantizations published	2	29
Smallest quantization	0.75 GiB	1.01 GiB
Q4_K_M	0.75 GiB	2.33 GiB
Licence	mit	—

KV cache by context

the term that decides long-context viability

Context	ultravox-v0_5-llama-3_2-1b	Qwen3-4B	Ratio
4,096	1.00 GiB	0.56 GiB	1.78×
8,192	2.00 GiB	1.13 GiB	1.78×
16,384	4.00 GiB	2.25 GiB	1.78×
32,768	8.00 GiB	4.50 GiB	1.78×
65,536	16.00 GiB	9.00 GiB	1.78×
131,072	32.00 GiB	18.00 GiB	1.78×