DeepSeek raises AI model prices as new V4 models arrive

DeepSeek is preparing to significantly increase the cost of accessing its artificial intelligence models, ending a period in which the Chinese AI company became known for offering relatively low-cost services. The changes will raise API prices for its latest models by more than four times during peak periods.

The company has announced new pricing for its DeepSeek V4 Pro and V4 Flash models, introducing separate rates for peak and off-peak usage. The changes are due to take effect on 16 August, with DeepSeek saying the new system is designed to use its computing resources more efficiently.

The move marks a notable shift for a company whose low prices have been a major part of its appeal to developers and businesses. DeepSeek gained international attention by offering capable AI models at prices significantly below those of many Western competitors, helping fuel wider competition in the AI market.

DeepSeek V4 Pro faces the biggest price increase

Under the new pricing structure, DeepSeek V4 Pro will cost US$3.96 for 1 million output tokens during peak hours. That represents a substantial increase from the current rate of US$0.87 per 1 million output tokens.

During off-peak periods, the same amount of output from V4 Pro will cost US$1.98. DeepSeek said the introduction of different peak and off-peak rates is intended to “allocate resources more reasonably”, suggesting that the company is attempting to manage demand for its latest models while controlling access to computing capacity.

DeepSeek V4 Flash is also becoming more expensive, although it remains cheaper than V4 Pro. The model will cost US$1.32 per 1 million output tokens during peak hours, compared with the current price of US$0.28. Its off-peak rate will be US$0.66 per 1 million output tokens.

The increases mean developers using DeepSeek’s API will need to reconsider the cost of running applications powered by the company’s models. Businesses with large amounts of AI-generated output could be particularly affected, as token costs can add up quickly across high-volume workloads.

Price cuts were originally meant to continue

The price increases are particularly notable because DeepSeek had previously indicated that its discounted API rates would remain in place. The current prices were introduced as a promotional offer scheduled to end on 31 May.

In May, DeepSeek announced that it would make the discounted pricing permanent, giving developers and companies more confidence that the lower rates could be used for longer-term projects. The company’s latest decision means that plan will not continue as originally expected.

The reversal highlights the growing pressure faced by AI companies as demand for computing resources increases. Running advanced AI models requires substantial processing power and infrastructure, and companies must balance competitive pricing with the cost of operating their services.

DeepSeek’s decision to introduce peak and off-peak rates also reflects a wider challenge in the AI industry. As more users rely on AI models for software development, content generation, research, and business applications, providers must manage periods of heavy demand without letting infrastructure costs rise unchecked.

DeepSeek remains cheaper than several rivals

Despite the increases, DeepSeek’s new models are still priced below several competing AI services. Moonshot’s Kimi K3, for example, costs US$15 per 1 million output tokens, putting it considerably above DeepSeek V4 Pro’s peak rate of US$3.96.

OpenAI’s most advanced model, GPT-5.6 Sol, is listed at US$30 per 1 million output tokens in the supplied pricing comparison. That makes DeepSeek V4 Pro substantially cheaper even during peak periods.

However, the comparison becomes less straightforward when you consider lower-cost models. OpenAI’s GPT-5.6 Luna is priced at US$1.20 per 1 million output tokens, which is below DeepSeek V4 Flash’s new peak rate of US$1.32. This means DeepSeek will no longer automatically offer the lowest price in every part of the market.

The changes still leave DeepSeek in a relatively competitive position, particularly for customers who can shift workloads to off-peak periods. V4 Flash will cost US$0.66 per 1 million output tokens outside peak hours, while V4 Pro will cost US$1.98.

For developers choosing an AI provider, the new prices are likely to make factors such as performance, reliability, usage limits and peak-hour demand just as important as headline token costs. DeepSeek’s price increase could therefore mark a new phase in the competition between AI providers, where low prices remain important, but sustainable operating costs are becoming a larger consideration.