OpenAI and Anthropic Cut Prices as Chinese AI Models Gain Share

US AI Labs Enter Price War

OpenAI and Anthropic have sharply reduced pricing on several models as cost-conscious enterprise customers increasingly test alternatives from Chinese AI developers, according to a Financial Times report.

OpenAI recently cut prices for GPT-5.6 Luna by 80%, dropping from $1 to $0.20 per million input tokens and from $6 to $1.20 per million output tokens. Anthropic launched Claude Opus 5 at half the price of its most capable model, Fable 5, and this week canceled a planned price increase for its Sonnet 5 model.

The moves have driven down average prices customers pay for leading US models by nearly 25% since mid-July, according to Silicon Data's token price index.

Corporate Customers Seek Cost Relief

The pricing pressure comes as both OpenAI and Anthropic shift enterprise customers from flat subscription fees to usage-based billing, where companies pay according to computational resources consumed. Some businesses have responded by capping AI usage or testing cheaper alternatives.

Major companies including DoorDash and Airbnb have begun experimenting with Chinese-made models to control costs, the Financial Times reported. Chinese labs including Moonshot and DeepSeek have released increasingly capable open models that can be freely downloaded and modified by developers, narrowing the performance gap with US offerings.

Mid-Tier Models Under Pressure

The latest price cuts apply primarily to mid-tier products rather than flagship models. Mantas Lukauskas, AI tech lead at website hosting provider Hostinger, told the Financial Times that prices for the most advanced models remain "flat to rising."

"The US labs have cut the middle and are defending the top," Lukauskas said, calling the pricing changes the "first real test" of whether OpenAI and Anthropic can protect margins on their most advanced offerings.

Benchmarking firm Artificial Analysis found that Anthropic's Opus 5 at medium effort settings delivered similar performance and cost per task to Moonshot's Kimi K3 at maximum effort. OpenAI's GPT-5.6 Luna at maximum effort performed comparably to DeepSeek's V4 Flash but cost nearly twice as much per task.

Why it matters

The price war arrives as OpenAI and Anthropic prepare for potential initial public offerings at trillion-dollar valuations, with investors demanding evidence that massive AI infrastructure spending can generate sustainable returns. Chinese competitors offering capable open models at lower costs threaten the pricing power US labs have relied on to justify their valuations, potentially forcing a fundamental shift in business models before either company goes public.

These details were first reported by Jamie John and Clara Murray for the Financial Times.