A price war is tearing through China’s artificial intelligence sector, one that began with token prices dropping to $0.10 per million and has since accelerated further, and the South China Morning Post’s latest analysis makes clear that it is deepening rather than stabilizing. As the capability gap between leading Chinese and American AI models narrows, competition has shifted from technical differentiation to pricing, leading to a race to the bottom that is rewriting the economics of AI deployment for enterprise customers worldwide.
The numbers are stark. DeepSeek’s V4 Flash model is currently priced at $0.14 per million input tokens and $0.28 per million output tokens. Its flagship V4 Pro model ranges from $0.435 to $1.74 per million input tokens and $0.87 to $3.48 per million output tokens, depending on the tier. Alibaba’s Qwen3.5-Flash is available at $0.10 per million input tokens and $0.40 per million output tokens, while the higher-performance Qwen3-Max is priced at $0.359 and $1.434, respectively.
Compare those figures to the American alternatives. OpenAI’s GPT-5.5 is priced at $5 per million input tokens and $30 per million output tokens. Anthropic’s Claude Sonnet 4.6 costs $3 for input and $15 for output; its premium Claude Opus 4.8 model is $5 and $25. The pricing differential between the leading Chinese and American models is not marginal, it is an order of magnitude or more, and it is growing.
The Mechanics of the Price War
The immediate trigger for the current round of price reductions was DeepSeek’s announcement of a permanent 75 percent cut to its V4-Pro API pricing. The move forced every other Chinese model developer to respond, and the competitive dynamic has been self-reinforcing: as prices fall, the addressable market for AI API services expands, increasing the volume of API calls, which in turn reduces per-unit costs, creating room for further price reductions. The cycle is familiar from the history of cloud computing, where a similar dynamic drove the commoditization of compute infrastructure over a decade.
What is different about the AI model price war is the speed at which it is unfolding and the degree to which it is being driven by capability convergence rather than purely by cost reduction. A year ago, the price differential between Chinese and American models was justified in part by a genuine capability gap, American frontier models were meaningfully more capable across a range of tasks. That gap has narrowed significantly. DeepSeek V4 and Alibaba’s Qwen3 series now match or exceed American models on many standard benchmarks, and for the majority of enterprise use cases, document processing, code generation, customer service automation, data extraction, the practical difference in output quality is difficult to detect.
When capability parity is achieved, price becomes the primary competitive variable. Chinese model developers, operating with lower infrastructure costs, a domestic talent pool that commands lower salaries than Silicon Valley, and state support that reduces capital costs, have a structural advantage in a price competition. The current pricing levels reflect that advantage being deployed aggressively.
Enterprise Customers as the Prize
The price war is fundamentally a competition for enterprise customers, the companies that will embed AI models into their products and workflows at scale and generate the sustained API revenue that justifies the enormous capital investment in model development. For a Chinese enterprise customer choosing between a domestic model at $0.14 per million tokens and an American model at $5 per million tokens, the economics of the decision are not close. For an international enterprise customer in Southeast Asia, the Middle East, or Africa, the calculation is similar, and the disruption to Anthropic’s Fable 5 service this week has made the American alternative less reliable as well as more expensive.
The consolidation that analysts have been predicting for China’s crowded AI model market is now beginning to take shape through pricing rather than through the kind of explicit market exits or mergers that characterized earlier technology cycles. Companies that cannot sustain the current pricing levels without burning through capital at an unsustainable rate will face a choice between raising more money, finding a strategic partner, or exiting the market. The companies with the deepest pockets, including DeepSeek, Alibaba, and ByteDance, are best positioned to outlast the competition, and the current pricing levels may be as much a strategic weapon as a reflection of genuine cost efficiency.
For the global AI industry, the Chinese price war is creating a new baseline expectation for what AI API services should cost. That expectation will be difficult to unwind, and it is already putting pressure on American model developers to justify their premium pricing with demonstrably superior capabilities. The race is no longer just about who can build the most powerful model, it is about who can build a powerful enough model at a price that the market will actually pay.