Price war at the top as OpenAI and Anthropic cut rates

Two leading U.S. labs have entered an open price competition. OpenAI reduced the fee for GPT-5.6 Luna by 80 percent, describing it as its “fastest and cheapest” model. Anthropic introduced Cloud Opus 5 at half the price of Claude 5, the most powerful model in its catalog to date. The moves come amid cheaper alternatives from Chinese rivals such as Moonshot and DeepSeek, prompting some American firms to switch to those options to trim expenses.

According to Silicon Data’s token-price index, the average price customers pay for models from the leading labs has fallen by almost a quarter since mid-July. Tokens, the data units processed by language models, form the basis of billing for most commercial offerings, so any discount flows directly to the final invoice. This signals a sharp shift: competition that once centered on performance now hinges on cost.

Chinese labs are releasing “open” models that can be downloaded for free, run locally, and customized, unlike closed-API services. This gives developers full control over infrastructure costs, eliminating usage-based charges. Performance gaps have narrowed to a negligible margin compared with the top U.S. models, making the Chinese alternatives viable for production use.

In the background, both OpenAI and Anthropic are preparing IPOs valued at $1 trillion each, and investors are demanding proof that massive infrastructure and research spending translates into returns. At the same time, the two companies are moving enterprise customers from flat-rate subscriptions to usage-based billing, where the bill grows with each token consumed. Companies such as DoorDash and Airbnb report that they have begun integrating Chinese models to curb bill spikes.

The rollout of Chinese releases has reduced the technical gap to a point where the premium no longer justifies the cost for many applications. In the U.S. market, there is growing concern that the billions spent on a technological edge will not secure customer loyalty when the cheaper alternative is “good enough.” The price war has only begun, and the winners will be those who can deliver high-level intelligence at a cost organizations are prepared to sustain over time.