Is dots.llm1.inst or DeepSeek-V4-Flash smaller?

Both publish quantizations; compare the table for the exact sizes.

Which uses less memory at long context?

DeepSeek-V4-Flash — its KV cache at 32K context is 492.2x smaller, because of how its attention layers cache.

Model comparison

dots.llm1.inst vs DeepSeek-V4-Flash

These two publish different quantization sets; the table below has the exact sizes. At long context the gap widens: DeepSeek-V4-Flash's KV cache at 32K is 492.2× smaller, which usually matters more than the difference in weights.

From the file· summed bytes, KV per layer

Side by side

	dots.llm1.inst	DeepSeek-V4-Flash
Parameters	143B	291B
Architecture	dots1	deepseek4
Layers	62	43
Native context	32,768	1,048,576
Mixture of experts	yes, 128 experts	yes, 256 experts
Quantizations published	46	12
Smallest quantization	44.04 GiB	76.87 GiB
Q4_K_M	88.00 GiB	—
Licence	mit	mit

KV cache by context

the term that decides long-context viability

Context	dots.llm1.inst	DeepSeek-V4-Flash	Ratio
4,096	3.88 GiB	0.06 GiB	61.52×
8,192	7.75 GiB	0.06 GiB	123.04×
16,384	15.50 GiB	0.06 GiB	246.08×
32,768	31.00 GiB	0.06 GiB	492.16×
65,536	62.00 GiB	0.06 GiB	984.31×
131,072	124.00 GiB	0.06 GiB	1968.62×