DeepSeek warned developers Thursday of a significant price increase across its API services, an about-face for a company charging $0.14 per million input tokens. Key Points: DeepSeek told use
DeepSeek warned developers Thursday of a significant price increase across its API services, an about-face for a company charging $0.14 per million input tokens.
Key Points:
- DeepSeek told users a significant API price rise is coming, without naming figures or an effective date.
- The company currently charges far below Western rivals, at $0.14 per million input tokens and $0.28 per million output tokens for V4 Flash.
- The warning follows record demand for V4-Flash-0731 and plans for a large data center in Inner Mongolia.
The Hangzhou-based company posted the notice on its developer platform, offering no figures, no percentages and no effective date for the change.
It said only that overall pricing would rise in the near future, with detailed plans to be advised later. Customers were told to plan their usage accordingly, a phrase that gave no indication of scale.
DeepSeek charges $0.14 per million input tokens and $0.28 per million output tokens for V4 Flash, rates that undercut most of the market by an order of magnitude. Moonshot AI asks $3 and $15 for the same volumes on Kimi K3, while Anthropic lists $10 and $50 for the same work on its Fable 5 service.
The warning arrived a week after the release of V4-Flash-0731, a 284-billion-parameter model that activates roughly 13 billion parameters at a time, a design choice that holds the cost of a typical task near three cents.
Also Read:Airbnb Stock Jumps 17% As CEO Says AI Resolves 45% Of Support Issues Without Human Help
AI developer Michael Guoquestioned the timing in a social media post on Thursday, arguing the company was inviting trouble by moving now.
He noted that American labs have closed much of the gap DeepSeek once enjoyed on both cost and capability. Meta's Muse Spark and OpenAI's GPT-5.6 Luna now compete directly for the same developer budgets.
Usage figures show how quickly the strain built after the model reached general availability. OpenRouter rankings put V4 Flash first for the week of Jul. 27 to Aug. 2, at 7.22 trillion tokens, and the coding platform OpenCodelogged 8 trillion tokens in a single day on Aug. 1.
Serving traffic on that scale consumes chips and electricity in quantities that bargain pricing struggles to cover for long. DeepSeek plans a large data center in Inner Mongolia as part of a push to secure a gigawatt of computing power for its own use.
Nvidia chief executive Jensen Huang has estimated that a build of that size would cost roughly $50 billion when equipped with the most advanced accelerators.
The increase marks DeepSeek's second pricing change in under a month, following the peak and off-peak rates it introduced in mid-July to ease pressure on its servers during the busiest hours.
Before that, a permanent discount on V4 in May pushed domestic rivals including ByteDance and Tencent to trim their own published rates.
Read Next:Google DeepMind Loses Four Founding-Era Leaders In A Single Day
