Chinese AI models not only have a price advantage, but they are also becoming more and more competitive. US AI giants are forced to reduce prices to retain customers.
According to a report by the British 'Financial Times' on August 14, as businesses struggle with rising costs associated with AI-generated invoices, more and more customers are looking for cheaper alternatives. Chinese AI companies such as DeepSeek and Moon Dark Face are taking advantage of this trend to expand their influence, seeking users both in Silicon Valley and Europe. This has also put pressure on American companies like OpenAI and Anthropic.
OpenAI recently announced that it has reduced the price of GPT-5.6 Luna by 80%. The price per million input tokens has been lowered from $1 to $0.2, while the output price has decreased from $6 to $1.2. Anthropic’s newly launched Claude Opus 5 charges $5 per million input and output tokens, which is only half of its flagship model Fable 5. This week, Anthropic also canceled its planned price increase for Sonnet 5, which was scheduled to take place in September.
Market data has reflected this change. The token price index of Silicon Data shows that since mid-July, the prices paid by customers for models from leading American AI companies have dropped by nearly a quarter.
The Financial Times states that American AI companies have traditionally relied on performance to compete. However, as China's capabilities in open-source models continue to improve, price has become an increasingly important factor in competition.
Business clients have begun to switch to cheaper options. As OpenAI and Anthropic transition some business clients from fixed subscriptions to pay-per-usage models, the AI bills for some companies have significantly increased. As a result, companies are forced to limit employees’ use of AI tools or test other models. DoorDash and Airbnb have indicated that they have already started using Chinese-based models to control costs.
An industry insider summarized the current strategy of American AI companies as: “Cut prices at the middle level, and defend the top position.”
In fact, the Chinese models not only have cost advantages, but the performance gap between them and top US products is also narrowing.
The evaluation agency Artificial Analysis found that under corresponding settings, Anthropic's Opus 5 and DeepSeek's V4 Flash perform similarly in terms of performance and single-task cost. OpenAI's GPT-5.6 Luna also performs comparable to DeepSeek's V4 Flash, but the cost to complete a single task is still nearly twice that of the latter.
