DeepSeek slashes prices by 75% for new flagship AI model

Chinese artificial intelligence (AI) startup DeepSeek has sharply cut prices for its newly launched flagship AI model, escalating the global race to offer more powerful systems at lower cost.

The Hangzhou-based company said developers will receive a 75% discount on its new DeepSeek-V4-Pro model until May 5, while prices for input cache hits across its full API lineup have been slashed to one-tenth of previous levels.

The aggressive pricing move comes just days after DeepSeek unveiled a preview of its long-awaited V4 family, a new generation of open-source large language models designed to run on Huawei chip technology rather than relying heavily on US suppliers.

DeepSeek said the V4 series includes a premium Pro version and a lighter, lower-cost Flash model, both aimed at developers building AI agents capable of carrying out more complex multi-step tasks than standard chatbots.

According to the company, DeepSeek-V4-Pro outperforms rival open-source systems on world knowledge benchmarks and trails only Google’s closed-source Gemini-Pro-3.1.

The launch also marks a significant milestone for China’s domestic AI ecosystem, as DeepSeek confirmed the model has been adapted for Huawei’s Ascend processors amid tightening US export restrictions on advanced semiconductors.

Analysts say the steep discounts are likely to intensify pressure on American AI leaders by pushing the industry further toward commoditised, low-cost model access instead of premium pricing.