Is DeepSeek-V4-Flash or step-3.5-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 206.9x smaller, because of how its attention layers cache.

Model comparison

DeepSeek-V4-Flash vs step-3.5-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 206.9× smaller, which usually matters more than the difference in weights.

From the file· summed bytes, KV per layer

Side by side

	DeepSeek-V4-Flash	step-3.5-flash
Parameters	291B	199B
Architecture	deepseek4	step35
Layers	43	45
Native context	1,048,576	262,144
Mixture of experts	yes, 256 experts	no
Quantizations published	12	30
Smallest quantization	76.87 GiB	38.47 GiB
Q4_K_M	—	111.70 GiB
Licence	mit	apache-2.0

KV cache by context

the term that decides long-context viability

Context	DeepSeek-V4-Flash	step-3.5-flash	Ratio
4,096	0.06 GiB	2.53 GiB	40.19×
8,192	0.06 GiB	4.03 GiB	64.00×
16,384	0.06 GiB	7.03 GiB	111.63×
32,768	0.06 GiB	13.03 GiB	206.88×
65,536	0.06 GiB	25.03 GiB	397.40×
131,072	0.06 GiB	49.03 GiB	778.42×