DeepSeek cuts V4-Pro prices by 75% for developers

DeepSeek said on April 27 it would give developers a 75% discount on its new DeepSeek-V4-Pro model and cut input cache hit prices across its API lineup to one-tenth of previous levels.

The move came after the company released a preview of its V4 model on April 25, with versions called Pro and Flash, and said the model was adapted for Huawei chips.

DeepSeek said V4-Pro beat other open-source models on world-knowledge benchmarks and trailed only Google’s closed-source Gemini-Pro-3.1.

The V4 models are designed for AI agents that handle more complex tasks than chatbots.

🔗 Source: Reuters

🧠 Food for thought

Implications, context, and why it matters.

DeepSeek cuts prices while pushing new hardware and facing U.S. distillation claims

The temporary discount can distract from an already low base price 1.
DeepSeek V4-Pro’s application programming interface (API) costs already sit far below Western rivals, at $3.48 per million output tokens versus $30 for OpenAI’s GPT-5.5 1.
V4 is presented as a new model tuned for Huawei’s Ascend processors, the Chinese company’s AI chips, rather than Nvidia hardware 2.
DeepSeek also faces allegations that it distilled U.S. models and used fraudulent accounts to reach Anthropic’s Claude model, according to a U.S. State Department cable and Anthropic, an AI company backed by Amazon and Google 2.

China’s AI setup adds pressure and could expand long-context uses

V4’s launch hit several Chinese AI stocks in Hong Kong trading on Friday. Zhipu, also known as Knowledge Atlas Technology, fell about 8% and MiniMax also dropped about 8% 3.
SMIC, China’s biggest contract chipmaker, rose 9% 3.
Investors treated the move as support for China’s effort to rely more on domestic AI hardware. V4 can run on homegrown chips such as Huawei’s Ascend processors, though the share of Huawei versus Nvidia hardware used in training remains unclear 3.
V4-Flash costs $0.14 per million input tokens and $0.28 per million output tokens 1. That pricing could lower the cost of one-million-token context windows and heavier AI agent work. It could support tasks such as reviewing entire codebases or regulatory filings in one request 1.