BREAKING
DeepSeek V4 API margins hit 70-80%
0
%
V4 gross margin
0
M
Annualized revenue
0
%
Overall model margin
V4-Pro vs V4-Flash
V4-Pro
MoE
●
1.6T total params
●
49B active
●
1M-token context
V4-Flash
MoE
●
284B total params
●
13B active
●
1M-token context
V4-Pro slashes inference cost
Per-token FLOPs
27
KV cache use
10
V4 API pricing per 1M tokens
Pro input
0.435
Pro output
0.87
Flash input
0.14
Flash output
0.28
Efficiency turns into profit
AI NEWS BLITZ
DeepSeek is reportedly earning software-like margins of 70 to 80 percent on its V4 model.