China-based AI developer DeepSeek announced a limited-time 75% price cut for developers subscribing to its newly revealed DeepSeek-V4-Pro model, with the offer running through May 5, the company said. In the same announcement, DeepSeek said it has lowered prices for input cache hits across its full DeepSeek API range to one-tenth of the previous rate.
The company released a preview of the V4 model on Friday, noting that the architecture has been adapted for compatibility with Huawei chip technology. DeepSeek positioned V4 as a two-tier product: a more capable - and higher-priced - Pro edition and a lighter, lower-cost Flash edition.
According to DeepSeek, tests place the Pro edition ahead of other open-source models on world-knowledge benchmarks, with the Pro model trailing only Google’s closed-source Gemini-Pro-3.1. The startup also highlighted that the V4 models are especially well suited for AI agent work - applications that can perform more complex, multi-step tasks than standard chatbots but which also demand greater computing resources.
The simultaneous pricing moves - a temporary steep discount on the Pro model and a broad reduction in input cache hit charges for the API lineup - indicate a notable short-term cost change for developers using DeepSeek’s services. The company did not provide additional details in the announcement about pricing beyond the discount period or long-term rate plans.
Context and implications
DeepSeek’s preview emphasizes performance comparisons and hardware adaptation without offering further product roadmaps or permanent pricing commitments. The V4 family’s dual-tier approach creates distinct choices for developers between higher-capacity and lower-cost options, while the cache hit price cut affects the billing of API-based usage across its product set.
Readers should note the company’s specific claims on benchmark placement and the stated suitability of V4 for agent-style AI work were presented by DeepSeek in its announcement.