Leading US artificial intelligence companies OpenAI and Anthropic are cutting the prices of some of their models as they compete to retain cost-conscious customers increasingly turning to cheaper alternatives from Chinese rivals.
The price war comes as rising AI bills encourage companies to reduce usage, impose spending limits and test lower-cost models. Chinese developers including Moonshot and DeepSeek are making inroads among users from Silicon Valley to Europe, increasing pressure on US companies that have invested heavily in maintaining their technological lead.
According to the Financial Times, OpenAI recently cut the price of GPT-5.6 Luna, described as its “fastest and most affordable model”, by 80 per cent. Anthropic has launched Claude Opus 5, presenting it as offering “frontier intelligence . . . at half the price” of Fable 5, its most capable model.
The reductions have contributed to a broader decline in the cost of using leading US AI models. Silicon Data’s token price index shows that prices customers pay for models from major US laboratories have fallen by almost a quarter since mid-July. Tokens are units of data processed by language models and are commonly used to calculate customer bills.
The shift marks a change for US companies whose proprietary, or “closed”, models have traditionally competed primarily on performance. Increasingly capable Chinese “open” models, which can be freely downloaded and modified by developers, have added to the pressure to make commercial AI systems more affordable.
The timing is particularly significant because OpenAI and Anthropic are reportedly preparing initial public offerings at trillion-dollar valuations, placing greater emphasis on whether the enormous sums being invested in artificial intelligence can ultimately generate sustainable returns.
Corporate customers are facing their own financial pressures as OpenAI and Anthropic move some enterprise users away from flat subscription arrangements towards usage-based billing. Under these systems, companies pay according to the computational resources consumed by their AI applications.
For businesses using AI extensively, the change can produce sharply higher bills as usage expands. Some have responded by imposing limits on AI consumption or experimenting with cheaper alternatives. Companies including DoorDash and Airbnb have said they have begun using Chinese-made models as they seek to control costs.
The move towards Chinese models has coincided with a series of releases from Chinese laboratories that have narrowed the performance gap with leading American systems. The development has raised concerns within the US technology industry that American AI developers could lose customers even while continuing to spend heavily to maintain their technological advantage.
The latest US price cuts focus on mid-tier models, making them more competitive with Chinese offerings. OpenAI has reduced GPT-5.6 Luna from $1 to $0.20 per million input tokens and from $6 to $1.20 per million output tokens. Anthropic has priced Opus 5 at $5 per million input tokens and $25 per million output tokens, half the price of Fable 5. The company also cancelled a planned September price increase for its Sonnet 5 model.
Headline token prices, however, do not provide a complete measure of the cost of using an AI system. More capable models can sometimes complete tasks using fewer tokens or fewer attempts, meaning a model with a higher nominal token price can ultimately be cheaper for a particular task.
Artificial Analysis, which benchmarks models in areas including mathematics, science, coding and reasoning, found that Anthropic’s Opus 5 at “medium” effort delivered similar performance and cost per task to Moonshot’s Kimi K3 at “max” effort. OpenAI’s GPT-5.6 Luna at “max” effort performed similarly to DeepSeek’s V4 Flash at “max”, but cost just under twice as much per task.
The comparison highlights another complication in the emerging AI price war. Models can operate at different “effort” settings, altering the amount of computing power used to answer a question and affecting both performance and the eventual cost of completing a task.
Anthropic and OpenAI declined to comment. A person close to Anthropic said the pricing of Opus 5 below the company’s flagship Fable 5 reflected how its “family of models is built” and was not connected to competitors.
Mantas Lukauskas, AI tech lead at website hosting provider Hostinger, said prices for the most advanced models remained “flat to rising”. He described the recent reductions as the “first real test” of whether companies such as Anthropic and OpenAI can protect the prices of their most sophisticated systems.
“The US labs have cut the middle and are defending the top,” he said, highlighting the emerging strategy among leading American AI companies: make mainstream models cheaper to retain customers while preserving premium pricing for their most powerful systems.
