DeepSeek told developers on 6 August that it plans to raise API prices "soon", and warned that the increase could be substantial. The notice, first carried by IT Home in Chinese, contains no rate card, no effective date and no reasoning. Customers were told to plan usage accordingly and wait for a separate official announcement.
What it costs today
DeepSeek-V4-Flash lists at $0.14 per million input tokens and $0.28 per million output, with lower rates for cached requests. That is roughly an order of magnitude under comparable Western frontier APIs, and it is the entire commercial proposition.
The second move in a month
In mid-July the company introduced peak and off-peak rates, charging more during busy hours. A blanket rise weeks later suggests time-shifting demand did not close the gap between what inference costs and what DeepSeek charges for it.
Reversing its own strategy
DeepSeek's releases in early 2025 forced price cuts across the industry and made "cheap Chinese model" a category. The company is now signalling that the position was not sustainable at its own scale — the same conclusion Western labs reached, arrived at from the opposite direction.
What is not said matters
Announcing a rise without a number is unusual, and the obvious readings — compute scarcity, memory prices tripling in 18 months, export-controlled accelerator supply — are inference, not disclosure. DeepSeek has confirmed only the direction.