Chinese company DeepSeek has announced a huge 75% discount for developers on its latest model, DeepSeek-V4-Pro, which it recently unveiled.
According to the company, this promotion will run until May 5, 2026, in a move aimed at luring developers away from expensive Western models.
The announcement also included a reduction in the costs of access to cache hit via its API to just 10% of its original price, significantly reducing the costs of running applications that rely on repetitive requests.
DeepSeek-V4: Power vs. Giants
The company launched a beta version of the V4 model last Friday, which has been optimized to run efficiently on Chinese Huawei chips . The model is available in two versions:
V4-Pro: The most powerful version designed for complex tasks, which the company claims outperforms most open-source models and comes second only to Google's Gemini 3.1 Pro model .
V4-Flash: The lightest, faster, task-oriented version that requires immediate response and very low cost.
"Astronomical" spreads
While the GPT-5.5 Pro model costs around $30, DeepSeek offers its V4-Pro model at a price equivalent to fractions of a dollar (about 6 yuan before deduction), which is more than 700x a difference in some cases.
DeepSeek emphasizes that the V4 models are specifically designed for AI Agents, software capable of performing autonomous and complex tasks that require high computing power and repeatability in requests, making lowering "cache" prices an irresistible competitive advantage for enterprises and startups.