Is Darwin-4B-Chimera or Llama-3.1-8B-Instruct smaller?

Darwin-4B-Chimera is smaller at Q4_K_M: 2.33 GiB against 4.58 GiB.

Which uses less memory at long context?

Darwin-4B-Chimera — its KV cache at 32K context is 3.1x smaller, because of how its attention layers cache.

Model comparison

Darwin-4B-Chimera vs Llama-3.1-8B-Instruct

At Q4_K_M, Darwin-4B-Chimera is the smaller download — 2,497,281,088 bytes against 4,920,739,168. At long context the gap widens: Darwin-4B-Chimera's KV cache at 32K is 3.1× smaller, which usually matters more than the difference in weights.

From the file· summed bytes, KV per layer

Side by side

	Darwin-4B-Chimera	Llama-3.1-8B-Instruct
Parameters	4.0B	8.0B
Architecture	qwen3	llama
Layers	36	32
Native context	40,960	131,072
Mixture of experts	no	no
Quantizations published	59	45
Smallest quantization	0.98 GiB	2.02 GiB
Q4_K_M	2.33 GiB	4.58 GiB
Licence	gemma	llama3.1

KV cache by context

the term that decides long-context viability

Context	Darwin-4B-Chimera	Llama-3.1-8B-Instruct	Ratio
4,096	0.56 GiB	0.50 GiB	1.13×
8,192	0.71 GiB	1.00 GiB	1.40×
16,384	0.90 GiB	2.00 GiB	2.22×
32,768	1.28 GiB	4.00 GiB	3.13×
65,536	2.03 GiB	8.00 GiB	3.95×
131,072	3.53 GiB	16.00 GiB	4.54×