Chinese AI startup DeepSeek has permanently reduced the pricing of its flagship V4 Pro model by 75%, a move that could intensify global competition in the AI industry and put additional pressure on rivals to lower their own prices.
In an announcement made on May 22, the company revealed that V4 Pro pricing has been cut to $0.003625 per million input tokens for cache hits, $0.435 per million input tokens for cache misses, and $0.87 per million output tokens. Tokens are the small units of text that AI systems process while generating responses.
DeepSeek introduced its V4 model lineup in April, including both the advanced Pro version and the lighter Flash variant. The company positioned the series as a cost-efficient solution offering a one-million-context-length capability along with strong reasoning, coding, and mathematical performance.
While DeepSeek did not directly explain the reason for the permanent price reduction, the decision comes as expectations grow around the broader rollout of Huawei’s Ascend 950 AI chips. DeepSeek had earlier highlighted these processors as an important factor in improving the efficiency and scalability of its V4 models.
When V4 was first launched, the company noted that the Pro version was priced nearly 12 times higher than the Flash model due to limited availability of high-end computing resources. DeepSeek had also indicated at the time that pricing could decline significantly once Huawei’s Ascend 950 supernode systems entered mass deployment later this year.
Huawei’s AI chip division has seen growing demand as US export restrictions continue limiting NVIDIA from selling its most advanced AI chips in China. However, Huawei still faces challenges in rapidly expanding production because restrictions on semiconductor manufacturing equipment remain in place.
DeepSeek’s aggressive pricing strategy is expected to increase competition across the AI sector, especially as companies worldwide compete to reduce inference costs and accelerate enterprise adoption of large language models.
Also Read: DeepSeek Releases V4 Pro, Taking on OpenAI and Anthropic in Key Benchmarks
