DeepSeek released a preview of its V4 AI model on April 24 and then said on Sunday it was cutting prices for its AI models, including V4, to draw enterprise clients, developers and agent-based users. [1, 2]
The company said input cache hits for API users would fall to one-tenth of the original level, with the change effective immediately and permanent. DeepSeek also said its latest V4 pricing was 97% lower than OpenAI products, while one report said developers got a 75% discount on the new V4 Pro. [2, 3]
DeepSeek said V4 was adapted for Huawei chip technology and that it had worked closely with Huawei on the model. Huawei said its Ascend supernode product line now supports DeepSeek-V4 series models. Huawei added in Chinese that “整个升腾超级节点产品线现在都支持DeepSeek-V4 系列模型。” [1]
The price cuts landed as the company pushed lower costs across its lineup, with minimum input pricing at about US$0.14 per million tokens and DeepSeek-V4-Pro input tokens priced at US$0.0036 per million. DeepSeek also compared that with OpenAI GPT-5.5 cached input pricing of US$0.5 per million cached input tokens, saying the ratio made GPT-5.5 about 32 times more expensive for a conversation. [2]
The timeline around the release and pricing was closely watched. DeepSeek's V4 preview came on April 24, one day after the White House accused China of large-scale theft of AI laboratory intellectual property, according to the report. [1]
DeepSeek's pricing change took effect immediately and was described as permanent. [2]
