Model FatigueLLMs under $1, measured — and who holds your data
Models

deepseek-v4-pro

DeepSeek V4 Pro

Made by DeepSeek. Sold by 5 providers, from $1.30 per million input tokens.

Maker
DeepSeek
Released
2026-04-24
Context
1,050K tokens
Max output
1,049K tokens
Input
text
Output
text
Open weights
Yes
Knowledge cutoff
2026-01

Where to buy it

ProviderInput $/MOutput $/MBlended*Holds your promptTraining
DeepInfra$1.30$2.60$1.63runs it 🇺🇸 United Statesno training
DeepSeek$1.32$3.96$1.98runs it 🇨🇳 China
Microsoft Azure$1.74$3.48$2.17runs it 🇺🇸 United Statesno training
nebius$1.75$3.50$2.19runs it 🇳🇱 Netherlandsno training
Alibaba Cloud$2.40$4.80$3.00runs it 🇨🇳 Chinano training

Blended is our calculation: (3 × input + output) ÷ 4, US dollars per million tokens — most calls read far more than they write. Tiers that are cheaper because they train on your data are not used for the “from” price.

Retirement dates

ProviderId on that providerShuts downSource
Microsoft AzureDeepSeek-V4-Pro2028-02-20provider page

Compare with every other model →