DeepSeek has permanently slashed the price of its flagship V4-Pro model, cutting DeepSeek V4-Pro API prices by 75% and taking the service to a quarter of its original level. The move is a clear bid to make premium AI usage cheaper and more attractive, especially in a market where compute access is often the real bottleneck, not model quality.
The Chinese AI startup said the V4-Pro API now costs 0.025-6 yuan per million tokens, or about $0.0035 to $0.83, depending on how it is used. That compares with the previous 0.1-24 yuan range. In plain English: DeepSeek just made its top-end model much easier to slot into products without the bill looking like a small national budget.
DeepSeek V4-Pro API prices drop to a quarter
Tokens are the chunks of text an AI model processes, so pricing per million tokens is the number developers actually care about. DeepSeek did not say whether the cut was tied to a bigger supply of Huawei Ascend 950 chips, but the timing is hard to ignore. Cheaper inference usually follows cheaper hardware, or at least a company pretending it found a magical coupon in the server room.
The company had previously said V4-Pro could cost up to 12 times more than the lighter Flash version because of limits in high-performance computing capacity. It also said prices should fall sharply after a mass rollout of Huawei Ascend 950 supernodes in the second half of the year. That earlier warning now reads less like a disclaimer and more like a roadmap.
Huawei chips are part of the story
DeepSeek’s pricing move also sits inside a bigger hardware squeeze. U.S. export controls have helped Huawei’s AI chip business by blocking Nvidia from selling its most advanced semiconductors in China, but restrictions on chip-making equipment have still limited Huawei’s ability to scale production of Ascend processors. In other words, the market is being reshaped by sanctions and supply constraints as much as by model performance.
For developers, the immediate winner is obvious:
- Lower costs
- More room to experiment
- A stronger case for using V4-Pro instead of downgrading to a cheaper model
For rivals, the pressure is just as obvious. Once one serious AI player starts discounting its flagship so aggressively, everyone else has to decide whether to follow or explain why their bills look so much healthier.
What developers will do next
The interesting question is whether this is a one-off pricing reset or the start of a broader race to the bottom in AI APIs. If DeepSeek can sustain these rates while scaling usage, it will strengthen the case for Chinese cloud and chip stacks as a cheaper alternative to the usual Nvidia-centered playbook. If not, the discount will be remembered as a sharp promo with a short shelf life.