US AI labs are chopping mid-tier model prices as Chinese rivals raid their cost-conscious customers with cheaper code.
According to the Financial Times, leading US AI labs, including OpenAI and Anthropic, are releasing cheaper models as they try to keep customers from scarpering to cut-price Chinese rivals.
The price scrap comes as swelling AI bills push companies to curb usage and hunt for cheaper models. That has helped Chinese developers, including Moonshot and DeepSeek, win users from Silicon Valley to Europe.
OpenAI recently said it was cutting prices for GPT-5.6 Luna, its “fastest and most affordable model”, by 80 per cent. Anthropic has launched Claude Opus 5, touting the system’s “frontier intelligence at half the price” of Fable 5, its most capable model.
The moves have cut the prices customers pay for models from leading US labs by almost a quarter since mid-July, according to Silicon Data’s token price index. Tokens are the chunks of data processed by language models and used to calculate many customer bills.
The cuts mark a shift for US AI outfits selling proprietary “closed” models, which have until now fought hardest on performance. More capable “open” Chinese models, which developers can freely download and tweak, have shoved more pressure on prices.
The moves come as OpenAI and Anthropic plot initial public offerings at trillion-dollar valuations. Investors are still looking for proof that the industry’s enormous AI spending can turn into returns, rather than smoke from a burning server rack.
Corporate AI users face cost pressure as Anthropic and OpenAI move some enterprise customers away from flat subscriptions and towards usage-based billing. Companies then pay based on the computational resources they consume.
Some businesses have responded to rising bills by capping AI usage or testing cheaper alternatives. DoorDash and Airbnb have said they have started using Chinese-made models to rein in costs.
That shift has arrived alongside a run of releases from Chinese labs narrowing the performance gap with leading US models. It has sparked concern in the US tech industry that American developers could lose customers while spending fortunes to keep their edge.
AI labs offer a range of models with different capabilities and prices. Costs vary further according to the model version and the “effort” settings used.
Customers are typically charged for input tokens, which measure data fed into a model, and output tokens, which measure what it generates in response.
The latest US cuts apply to mid-tier products, making them more competitive with Chinese offerings. OpenAI cut GPT-5.6 Luna from $1 to $0.20 per million input tokens and from $6 to $1.20 per million output tokens.
Anthropic launched Opus 5 at $5 per mn input tokens and $25 per mn output tokens, half the price of its Fable 5 model. This week, the company scrapped a planned Sonnet 5 price increase scheduled for September.
Headline token prices do not give a clean comparison between AI models. More capable models can sometimes complete a task with fewer tokens or attempts.
That means a model that looks pricier on headline token rates can end up costing less. This is the sort of AI pricing soup vendors enjoy because it makes simple maths look naive.
Most models can run at different “effort” settings, varying the computing power used to answer a question. That can affect performance and the final cost of completing a task.
Artificial Analysis, which benchmarks models across maths, science, coding and reasoning, found 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 less than twice as much per task.
