AI Model Pricing Drops as US Labs Compete With Chinese AI Providers

AI companies offer a wide range of models with different capabilities, pricing structures, and performance levels. Costs also vary by model version and the selected “effort” setting. Most customers pay separately for input tokens, which measure the data sent to an AI model, and output tokens, which measure the response it generates.

Recent price cuts from US-based AI labs have focused on mid-range models, making them more competitive with offerings from Chinese AI companies. OpenAI, for example, cut the price of GPT-5.6 Luna from $1 to $0.20 per million input tokens and from $6 to $1.20 per million output tokens.

Anthropic introduced Opus 5 at $5 per million input tokens and $25 per million output tokens—roughly half the price of its Fable 5 model. The company also canceled plans to increase pricing for its Sonet 5 models, which had been scheduled to rise in September.

However, comparing AI models based solely on advertised token prices can be misleading. A more capable model may require fewer tokens or complete a task in fewer attempts. As a result, an AI model with a higher per-token price may ultimately be more affordable for certain workloads.

Many AI models also offer adjustable “effort” settings. These controls determine how much computing power a model uses to solve a problem, affecting response quality, speed, token usage, and the total cost of completing a task.

Artificial Analysis, which evaluates AI models across areas including mathematics, science, coding, and reasoning, found that Anthropic’s Opus 5 delivered performance and per-task costs comparable to Moonshot’s Kim K3 when Opus 5 was used at “medium” effort and Kim K3 at “maximum” effort. OpenAI’s GPT-5.6 Luna at its “max” setting performed similarly to DeepSeek’s V4 Flash at maximum effort, but cost just under twice as much per task.

Anthropic and OpenAI declined to comment.

A person familiar with Anthropic’s strategy said Opus 5’s lower price compared with the flagship Fable 5 reflects how the company structures its model family and is not related to competitor pricing.

Mantas Lukauskas, head of AI technology at website hosting provider Hostinger, said pricing for the most advanced AI models has remained “flat to rising.” Hostinger has used large language models since 2020.

Lukauskas added that the latest AI price changes represent the “first real test” of whether companies such as Anthropic and OpenAI can maintain the high costs associated with developing cutting-edge models. He said US AI labs appear to be reducing prices in the middle of the market while continuing to protect premium pricing for their most advanced systems.

© 2025 Financial Times Company. Unauthorized reproduction prohibited. May not be redistributed, copied or modified in any way.