Chinese AI models pass 30% of US token usage on OpenRouter

Chinese artificial-intelligence models now account for more than 30% of the tokens that US companies run through the model marketplace OpenRouter, a sign that cost is reshaping enterprise AI choices.

That share has held above 30% every week since February 8, 2026 and peaked at 46%, up from an average of about 11% over the prior 12 months, according to OpenRouter usage data. In the first half of 2025 it was just 4.5%.

Price is the driver. Chinese open-weight models, including DeepSeek’s and Z.ai’s GLM family, cost 60% to 90% less than comparable US proprietary models. GLM 5.2 is priced at roughly one-fifth of Anthropic’s Opus 4.8 while scoring within a percentage point of it on agentic benchmarks, according to figures attributed to the developers.

“The price and performance gap is going to push companies toward whatever AI model gives them the best output for the dollar,” said Kevin Laird of Saltwater Interactive.

The capability trade-off is real: Chinese frontier models are estimated to trail US leaders by roughly six to nine months. The benchmark and pricing claims come from the model developers and have not been independently verified.