A new pricing battle is unfolding across the artificial intelligence industry, with OpenAI and Anthropic sharply reducing costs for some of their most widely used models while Chinese developers such as Moonshot and DeepSeek attract users from Silicon Valley to Europe.
A new pricing battle is unfolding across the artificial intelligence industry, with OpenAI and Anthropic sharply reducing costs for some of their most widely used models while Chinese developers such as Moonshot and DeepSeek attract users from Silicon Valley to Europe.
OpenAI said it was cutting the price of GPT-5.6 Luna, described by the company as its fastest and most affordable model, by 80 percent. Anthropic unveiled Claude Opus 5, promoting the system’s ‘frontier intelligence… at half the price’ of Fable 5, its most capable model.
Together, those moves have reduced what 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 data units processed by language models and often determine how much a client is billed.
The reductions mark a departure for American AI groups that build proprietary closed models, which until now have competed mainly on performance. The arrival of increasingly capable open Chinese models, which developers can download and modify freely, has added to the pricing pressure.
These changes come as OpenAI and Anthropic prepare for initial public offerings at valuations in the trillions of dollars while investors look for signs that the industry’s enormous AI spending can deliver returns.
Corporate AI users, meanwhile, face their own cost squeeze. Anthropic and OpenAI have been moving some enterprise customers away from flat subscriptions and toward billing tied to usage, under which companies pay according to the computing resources they consume.
Some businesses have responded by placing limits on AI use or trying cheaper alternatives. DoorDash and Airbnb, for example, have said they began using models from Chinese developers to help control costs.
That shift has unfolded alongside a series of releases from Chinese labs that have closed the performance gap with leading US models, raising concern among American tech companies that they could lose customers even as they spend heavily to stay ahead.
AI labs sell a range of models with different capabilities and prices, and costs vary further based on the model version and the effort settings a customer selects. Most providers charge for input tokens, which measure the data fed into a model, and output tokens, which measure what the model generates.
The latest cuts from US labs target middle tier products and make them more competitive with Chinese alternatives. OpenAI 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 set Opus 5 at $5 per million input tokens and $25 per million output tokens, half the price of its Fable 5 model. This week the company called off a planned increase in prices for its Sonnet 5 model that had been scheduled to take effect in September.
Headline token prices do not offer a simple comparison between models. More capable systems can sometimes complete a task with fewer tokens or fewer attempts, so a model that looks more expensive on paper can end up costing less. Most models also run at different effort settings that change computing power and therefore affect both performance and the final cost of a task.
Artificial Analysis, which benchmarks models in areas including math, science, coding and reasoning, found that Anthropic’s Opus 5 at medium effort delivered performance and cost per task similar to Moonshot’s Kimi K3 at maximum effort. OpenAI’s GPT-5.6 Luna at maximum effort performed like DeepSeek’s V4 Flash at maximum effort, but cost just under twice as much per task.
Anthropic and OpenAI declined to comment. A person close to Anthropic said the pricing of Opus 5 below its flagship Fable 5 reflects how the startup’s model family is built and has no connection to competitors.
Mantas Lukauskas, AI tech lead at website hosting provider Hostinger, which has used large language models since 2020, noted that prices for the very best models were flat to rising. He added that the recent changes are the first real test of whether companies such as Anthropic and OpenAI can protect the cost of their most advanced offerings. ‘The US labs have cut the middle and are defending the top,’ he said.
The latest round of cuts shows how quickly competitive pressure from open Chinese models is reshaping the economics of the AI industry, even as the biggest US labs try to defend their premium products.
