OpenAI’s AI Price War Heats Up as DeepSeek and Moonshot Push Back
The global AI race is moving faster than ever, and pricing has become the newest battleground. Shortly after OpenAI dramatically reduced the cost of its GPT-5.6 Luna model, two major Chinese AI players, DeepSeek and Moonshot, responded with moves that could intensify competition across the entire artificial intelligence industry.
OpenAI recently cut prices for GPT-5.6 Luna by as much as 80 percent. Input tokens now cost $0.20 per 1 million tokens, down from $1, while output tokens dropped to $1.20 per 1 million tokens from the previous $6 rate. The company said the reduction was made possible by improvements in model architecture and operational efficiency.
Still, many industry watchers viewed the sharp discount as more than a routine pricing update. It looked like the beginning of a direct AI pricing battle, especially against fast-growing Chinese labs that have become known for delivering powerful models at aggressive prices.
DeepSeek wasted little time answering.
The company introduced an updated version of its Flash-class model, known as DeepSeek V4 Flash 0731. The model reportedly uses 284 billion parameters, yet its performance is being compared with much larger frontier models, including systems believed to rely on multi-trillion-parameter architectures.
What makes DeepSeek’s move especially significant is its pricing. V4 Flash 0731 is being offered at $0.14 per 1 million input tokens and $0.28 per 1 million output tokens. That undercuts OpenAI’s newly reduced rates by a wide margin, especially on output costs, where the price difference is substantial.
For developers, startups, and enterprises building AI-powered products, these numbers matter. Token pricing directly affects the cost of running chatbots, coding assistants, automation tools, research agents, and large-scale AI applications. Lower prices can make advanced AI more accessible, particularly for companies that need to process large volumes of text.
DeepSeek’s latest release also reinforces a broader trend: Chinese AI labs are becoming increasingly competitive not only in model capability but also in cost efficiency. If a smaller model can deliver performance close to larger competitors while charging significantly less, it could reshape how businesses choose AI providers.
At the same time, Moonshot is reportedly increasing its computing power in a major way. The company behind the Kimi K3 model has secured access to a large cluster of 20,000 NVIDIA H200 GPUs through Alibaba, according to reports. This kind of hardware capacity can significantly expand training workloads and help accelerate the development of more advanced AI models.
The H200 is one of NVIDIA’s high-end AI accelerators, designed for demanding workloads such as large language model training and inference. A cluster of this scale gives Moonshot far more room to experiment, train, refine, and scale its AI systems.
Moonshot has also recently found itself in the middle of growing tensions between the United States and China over artificial intelligence development. Some U.S. officials and AI industry figures have accused the company of using model distillation techniques related to Western frontier AI systems. Moonshot has been linked to discussions around whether Chinese firms are using overseas access to advanced NVIDIA hardware to improve their models despite tightening U.S. export controls.
There have also been allegations that Moonshot may have had access to advanced NVIDIA GB300 servers, including through third-party locations such as Thailand. These claims remain part of the wider debate over how Chinese AI companies are acquiring compute resources and developing increasingly capable models.
The bigger picture is clear: the AI industry is entering a new phase where price, performance, and access to computing power are all colliding. OpenAI’s steep price cuts may have been intended to strengthen its market position, but DeepSeek’s immediate response shows that Chinese competitors are prepared to fight aggressively on cost.
For users and businesses, this competition could be good news. Lower AI model pricing may reduce operating costs, encourage more experimentation, and speed up adoption across industries. For AI companies, however, the pressure is rising. A model is no longer judged only by benchmark scores. It must also be affordable, efficient, scalable, and easy to integrate.
DeepSeek’s V4 Flash 0731 and Moonshot’s expanded GPU access suggest that China’s AI labs are not slowing down. If anything, they are becoming more strategic, more cost-focused, and more ambitious.
OpenAI may have started the latest round of AI price competition, but DeepSeek and Moonshot have made it clear that the battle is only beginning.
