Previously, when DeepSeek unveiled its V4 Pro model, the hefty price tag drew criticism from developers. The company clarified that the high pricing was mainly due to the constraints of high-end computing power and pledged to revise prices once computing capacity was enhanced. To placate developers, DeepSeek initiated a limited-time discount campaign. On April 24, the preview version of the V4 series was launched, open-sourcing two models—Pro and Flash—that support ultra-long contexts of up to one million tokens. Shortly thereafter, V4-Pro was offered at a 75% discount for a limited period, and the permanent price reduction was applied to API input cache hit prices across all models, bringing them down to one-tenth of their initial launch price. The input cache hit price per million tokens for V4-Flash is now just 0.02 yuan, while for V4-Pro, it is 0.025 yuan, setting a new global benchmark for large model pricing. Additionally, DeepSeek has successfully adapted V4 to work with eight domestic AI chip brands, including Huawei Ascend and Cambricon, as well as NVIDIA, achieving parity between domestic chips and NVIDIA GPUs. With enhanced computing power and completed chip adaptations, DeepSeek's API pricing is anticipated to decrease even further.