Recently, DeepSeek officially announced that the API of its DeepSeek-V4-Pro model will no longer return to the original price after the limited-time promotion ends on May 31, 2026, but will instead be permanently discounted, directly adjusted to one quarter of the original pricing. Previously, the model launched a full series API pricing adjustment on April 26, reducing the input cache hit price to one tenth of the launch price, and adding a limited-time 2.5 discount for the V4-Pro promotion. This strategic adjustment means that this highly competitive price will become the norm.

QQ20260525-085025.jpg

After the adjustment, the API pricing of DeepSeek-V4-Pro has set a new global low for large models. Specifically, the price for 1 million tokens of input (cache hit) is only 0.025 yuan, the input (cache miss) price is 3 yuan, and the output price is 6 yuan. This level of price reduction not only demonstrates DeepSeek's technical capabilities in model architecture optimization and computing efficiency, but also disrupts the existing commercial pricing system through extreme cost control.

Currently, the large model industry is accelerating from "technical emergence" to "commercial implementation," and enterprise customers are increasingly sensitive to computing costs and calling efficiency. By turning short-term promotions into permanent significant price reductions, DeepSeek is expected to further lower the barriers for developers and enterprises to build AI applications, giving rise to more frequent and high-throughput scenario-based implementations. This move not only solidifies its own market share but also forces the entire generative AI industry to re-examine business models, pushing global large model API pricing toward a more inclusive and rational range.