In August 2026, the artificial intelligence market is undergoing an unprecedented transformation. Leading American tech giants, OpenAI and Anthropic, are forced to revise their pricing policies, entering a fierce price war. The reason for this move is the aggressive expansion of Chinese developers, who have offered corporate clients powerful alternatives at significantly lower prices. In a climate where client companies are looking for ways to optimize costs, American leaders risk losing their positions if they cannot offer competitive terms.
Aggressive Price Cuts: GPT-5.6 Luna and Claude Opus 5
The situation in the market intensified in mid-summer 2026. OpenAI, striving to retain its client base, announced a sharp reduction in the cost of servicing its GPT-5.6 Luna model — currently the fastest and most accessible. The entry cost was reduced by 80%: from $1 to $0.20 per 1 million input tokens. A similar reduction affected output data as well — from $6 to $1.20 per million tokens. This is a strategic move aimed at making American technologies accessible to the mass market segment.
Anthropic made a counter-move as well. The developers released the Claude Opus 5 model, positioned as a powerful alternative to flagship solutions, but at half the price of their own Fable 5 model. Opus 5 rates are set at $5 for input and $25 for output tokens. Furthermore, Anthropic canceled the planned September price increase for its mid-tier Sonnet 5 model, acknowledging the need to stabilize costs for partners.
The Chinese Challenge and Customer Exodus
The catalyst for this price race was the success of Chinese AI models. Until recently, American labs competed exclusively in terms of performance and quality, ignoring the price factor. However, Chinese developers, such as Moonshot and DeepSeek, managed to significantly narrow the technological gap, offering systems with comparable characteristics but at a price that became critically attractive for business.
The result was an exodus of major corporate clients. Giants like DoorDash and Airbnb have already switched to Chinese models, seeking to reduce operating costs. Previously, shifting some clients to token-based consumption payments led to a sharp spike in their costs, forcing many companies to seek alternatives. Now, with Chinese systems offering high quality at low cost, American companies are forced to dump prices to regain market interest.
Complex Math: Token Price vs. Task Cost
However, a simple comparison of prices per million tokens can be misleading. Experts note that more powerful models are often capable of solving tasks with less resource consumption, which ultimately makes their use more economically beneficial, despite high rates. For example, the Anthropic Opus 5 model on medium settings demonstrates results comparable to the Chinese Moonshot Kimi K3 on maximum settings. At the same time, OpenAI GPT-5.6 Luna at maximum power performs at the level of DeepSeek V4 Flash, but costs almost twice as little per task.
Thus, the price war is not just a reduction in rates, but a complex struggle for efficiency. American companies are trying to prove that their models, even when cheaper, maintain an advantage in quality and speed of executing complex tasks, which is more profitable for business in the long run.
Financial Ambitions and IPO
In addition to retaining the client base, OpenAI and Anthropic are driven by ambitious plans to go public. Both companies are aiming for a valuation of at least $1 trillion. To attract investors, they need to demonstrate that huge investments in AI infrastructure are capable of generating real profits. Price dumping may seem like a risky move, but it is necessary to maintain market share and prove business scalability before the initial public offering.
