DeepSeek’s bargain AI pricing is getting more expensive

DeepSeek will sharply increase prices for its V4 AI models and introduce different rates for busy and quieter periods, with the changes taking effect on Aug. 16.

Saritha Rai reports for Bloomberg that the Hangzhou-based company is moving away from a single low price as it seeks to manage demand for its computing capacity.

For DeepSeek-V4-Flash, the price for generating one million output tokens will rise to $1.32 during peak hours. During off-peak periods, the rate will be $0.66. The model previously cost $0.28 per million output tokens.

Output tokens are the units of text an AI model generates in a response. They are a central cost factor for businesses that use AI tools at scale, such as for producing large volumes of content, summaries or customer-service replies.

Higher rates for V4-Pro

DeepSeek will also raise prices for its more capable V4-Pro model. It will charge $3.96 per million output tokens at peak times and $1.98 outside those periods. The previous price was $0.87 per million tokens.

The company says it is adjusting prices to allocate resources more efficiently. By making peak periods more expensive, it aims to encourage developers and corporate customers to move non-urgent tasks to less congested hours.

Even after the increase, DeepSeek remains less expensive than some major competitors. Rai notes that Anthropic charges $50 per million output tokens for its Fable 5 service.

DeepSeek’s unusually low prices have fuelled debate about whether cheaper Chinese models could pressure the margins of established AI providers. The company is also raising funds and preparing for a possible initial public offering, according to Bloomberg. That could increase pressure on founder Liang Wenfeng to balance growth, investor expectations and the high cost of computing infrastructure.