Is granite-4.0-1b or Llama-3.2-3B-Instruct smaller?

granite-4.0-1b is smaller at Q4_K_M: 0.95 GiB against 1.88 GiB.

Which uses less memory at long context?

granite-4.0-1b — its KV cache at 32K context is 1.4x smaller, because of how its attention layers cache.

Model comparison

granite-4.0-1b vs Llama-3.2-3B-Instruct

At Q4_K_M, granite-4.0-1b is the smaller download — 1,023,645,440 bytes against 2,019,377,440. At long context the gap widens: granite-4.0-1b's KV cache at 32K is 1.4× smaller, which usually matters more than the difference in weights.

From the file· summed bytes, KV per layer

Side by side

	granite-4.0-1b	Llama-3.2-3B-Instruct
Parameters	1.6B	3.2B
Architecture	granite	llama
Layers	40	28
Native context	131,072	131,072
Mixture of experts	no	no
Quantizations published	35	39
Smallest quantization	0.44 GiB	0.85 GiB
Q4_K_M	0.95 GiB	1.88 GiB
Licence	apache-2.0	—

KV cache by context

the term that decides long-context viability

Context	granite-4.0-1b	Llama-3.2-3B-Instruct	Ratio
4,096	0.31 GiB	0.44 GiB	1.40×
8,192	0.63 GiB	0.88 GiB	1.40×
16,384	1.25 GiB	1.75 GiB	1.40×
32,768	2.50 GiB	3.50 GiB	1.40×
65,536	5.00 GiB	7.00 GiB	1.40×
131,072	10.00 GiB	14.00 GiB	1.40×