Chinese AI Models Are Winning Over OpenAI and Anthropic Customers on Price

US companies are increasingly ditching OpenAI and Anthropic models in favor of cheaper Chinese alternatives. The shift comes as prices for flagship American models keep rising and the Trump administration imposes restrictions on the availability of some of them abroad.

Scale of the Shift

Data from OpenRouter, a platform that brokers access to language models for thousands of companies, shows a trend that's hard to ignore. The share of queries routed to Chinese models instead of American ones has been rising since February and hasn't reversed, even though just a year ago it was a marginal part of the market.

What matters is how companies are starting to split their workloads. When a project doesn't require the best available model, teams increasingly route it to the cheapest model that gets the job done well enough. In that game, Chinese models are winning, since the price gap spans an order of magnitude while the quality gap has all but disappeared.

GLM-5.2 Takes the Lead

The most talked-about name in this shift is GLM-5.2, a model from Beijing-based Z.ai (also known as Zhipu), released in June 2026. On the Artificial Analysis Intelligence Index, the model ranked fourth globally, ahead of competing American and Chinese models, including DeepSeek V4 Pro and Kimi K2.6. On the Code Arena leaderboard for front-end coding, the model's Max variant placed second, ahead of Claude Opus variants.

Price remains the main selling point, though. Z.ai priced GLM-5.2 at $1.40 per million input tokens and $4.40 per million output tokens. That, market analysts note, amounts to a fifth to a seventh of what it costs to use Claude Opus or GPT-5.5 per output token.

The model is as good as what's currently available from OpenAI and Anthropic - David Sacks, former White House AI advisor

Sanctions and Chips

The model was trained and run exclusively on Chinese Huawei Ascend 910B processors, using roughly 100,000 units. That shows Chinese AI developers can build competitive systems despite export restrictions on Nvidia's newest chips, which have for years been meant to limit China's access to Western technology. GLM-5.2 was released under the MIT license, meaning no commercial restrictions, letting companies run it locally for free.

Political decisions in the United States have also contributed to the market shift. Government restrictions have suspended some features of Anthropic's Claude Fable 5 and Mythos 5 models for users abroad, while OpenAI is rolling out its newest model gradually. Against that backdrop, a free, open model that matches paid competitors on quality has become an obvious choice for many companies.

What It Means for Polish Companies

For Polish businesses using language model APIs, this is a signal to check whether a given task actually requires the most expensive model available. A model router, one that sends simple queries to cheaper engines and complex ones to pricier engines, is no longer a technical curiosity but a standard cost-saving practice. At the same time, companies handling sensitive data will need to reckon with questions about whether Chinese models comply with EU data protection rules and the upcoming EU AI Act.

Price pressure could also force OpenAI and Anthropic to revisit their own pricing. Both companies have so far defended the premium positioning of their flagship models, citing higher quality and safety. If the quality gap keeps narrowing, it will get harder to justify bills several times higher for customers doing the same tasks.

Analysts note this is only the start of the cycle. Other Chinese labs, including DeepSeek and Moonshot AI (maker of the Kimi series), are preparing their own model updates later this year. If the pace of releases holds, American providers will have to compete not just on quality but on price, something they haven't had to do until now.

Sources: TechRepublic (techrepublic.com), The Next Web (thenextweb.com), Crypto Briefing (cryptobriefing.com)