Chinese AI models have secured over 30 percent of traffic on the aggregation platform OpenRouter every week since early February, a sharp rise from an average of 11 percent last year. This shift occurs because providers such as DeepSeek and Z.ai offer significantly lower per-token pricing than systems from OpenAI and Anthropic. The startup Lindy recently moved all its operations from Claude to DeepSeek, with CEO Flo Crivello stating the change saves millions of dollars. OpenRouter employee Justin Summerville notes that these Chinese open-source models run 60 to 90 percent cheaper than their American counterparts. As per-token costs for US providers continue to climb, the financial incentive for companies to switch becomes increasingly difficult to ignore.
Despite the cost advantage, experts caution that a performance gap still exists between the two groups. Kyle Chan at the Brookings Institution estimates the difference in capability is roughly six to nine months. A report from the Center for AI Standards and Innovation published in May found Chinese models trailed leading US systems by about eight months across cybersecurity, software development, math, science, and abstract reasoning. The widening cost gap does not immediately erase the quality disparity, forcing buyers to weigh immediate savings against potential accuracy risks.
- Chinese models accounted for 46 percent of OpenRouter traffic at peak times this year.
- The performance gap between Chinese and US models is estimated at six to nine months.
- Chinese open-source models run 60 to 90 percent cheaper than equivalent US systems.
