DeepSeek Permanently Slashes V4-Pro AI Model Price by 75% and Boosts API Performance

The Chinese artificial intelligence startup DeepSeek has made a significant strategic move by announcing a permanent 75% price cut for its high-end DeepSeek-V4-Pro AI model. This decision, which will take effect after the current promotional period concludes on May 31, 2026, positions the model's pricing at just one-quarter of its original listed cost. The aggressive reduction is a clear signal of DeepSeek's intent to intensify the ongoing AI price war, particularly against major American counterparts, and to broaden its appeal to a wider base of developers and enterprise users. Beyond the dramatic price adjustment, DeepSeek has also rolled out substantial enhancements to its API services. On May 23, 2026, the company officially confirmed that its API has undergone a new round of optimization and upgrades, resulting in significantly faster output speeds and improved service stability. A key aspect of these API improvements is the expanded service capacity. The platform will now natively support 500 concurrent online connections, a substantial increase designed to meet the growing demands of enterprise-level applications. For businesses requiring even greater concurrency, DeepSeek has opened an online application channel, allowing them to request higher limits. This combined strategy of extreme cost-effectiveness and robust technical performance is designed to capture a larger share of the fiercely competitive large-model AI market. The DeepSeek-V4-Pro model itself is not a lightweight offering; it was engineered as a high-end reasoning system. Originally, the model was priced at approximately $1.74 per million uncached input tokens and $3.48 per million output tokens. With the permanent price cut, these rates will fall to about $0.435 and $0.87, respectively. Furthermore, pricing for input cache-hit, which is vital for repeated prompts and long-running AI agents, has been reduced even more drastically, in some cases to one-tenth of previous costs. These reductions could translate into millions of dollars in annual savings for enterprise developers processing billions of tokens monthly. This strategic shift is partly enabled by infrastructure developments. DeepSeek's V4 series models are optimized to run on Huawei's Ascend AI accelerators, moving away from a primary reliance on Nvidia hardware. The increasing availability of Huawei's Ascend 950 and 950PR AI supernode systems likely provided DeepSeek with the confidence to sustain these permanently lower prices. This move also highlights a broader trend of coordinated software and hardware development within the Chinese AI ecosystem.