Chinese AI models are becoming harder for US companies to ignore. The reason is straightforward: many businesses are comparing performance against cost, and lower pricing is changing which models they are willing to run at scale.
According to CNBC, models from companies such as DeepSeek and Z.ai are increasingly viewed as competitive alternatives to systems from OpenAI and Anthropic. On OpenRouter, that shift is already visible in traffic patterns.
Cost is moving model choice
AI model selection is not only a technical decision. For companies that send large amounts of traffic through language models, per-token pricing can quickly become a major operating cost.
That is where Chinese AI models are gaining attention. OpenRouter employee Justin Summerville said Chinese open-source models run 60 to 90 percent cheaper. For companies processing high volumes of prompts and responses, that kind of gap can alter procurement decisions, product margins, and experimentation budgets.
The pressure is sharper because per-token pricing from US providers keeps climbing, while models from DeepSeek and Z.ai are described as competitive. A model does not need to be seen as identical to the most advanced US system to win workloads. It needs to be good enough for the task and cheap enough to justify switching.
OpenRouter traffic shows the shift
OpenRouter has become a useful signal for how developers and companies distribute AI traffic across model providers. On that platform, Chinese models have accounted for over 30 percent of traffic every week since February 8.
At times, the share has reached 46 percent. That is a sharp change from last year, when the average was just 11 percent.
Those figures suggest that Chinese AI models are not only being tested. They are being used repeatedly enough to hold a meaningful share of weekly traffic. The movement is especially notable because it is happening among users who can compare multiple models and route usage based on price, performance, and availability.
The result is a more competitive market for AI infrastructure. If users can switch between models more easily, pricing becomes more visible. Providers then compete not only on benchmark strength or brand trust, but also on whether their models make economic sense for everyday production workloads.
Lindy shows what a full switch can mean
The startup Lindy offers one concrete example. The company shifted all of its traffic from Anthropic's Claude to DeepSeek.
CEO Flo Crivello said the switch saves millions. That single example captures why the cost gap matters: for companies with enough usage, model pricing can become a board-level issue rather than a small engineering preference.
Moving all traffic from one model family to another is a significant decision. It implies that the cheaper model was acceptable for the company's needs, at least across the workloads Lindy chose to move. The source does not specify which tasks were involved, but the financial point is clear: the difference in provider costs was large enough to drive a major migration.
The capability gap is narrowing, but still present
Cost is not the only factor. Companies also need models that can perform reliably across the types of work they actually use them for. That is why estimates of the gap between Chinese and US models matter.
Kyle Chan at the Brookings Institution puts the gap between Chinese and US models at six to nine months. That estimate is close to the assessment from the Center for AI Standards and Innovation (CAISI).
CAISI published a report in May finding that Chinese AI models trail leading US models by about eight months. The assessment covered cybersecurity, software development, math, science, and abstract reasoning.
This matters because the market does not wait for perfect parity. If Chinese models are behind but cheaper by a wide margin, companies may still choose them for many uses. The tradeoff depends on how sensitive a task is to the leading edge of model performance.
What this means for the AI market
The OpenRouter numbers point to a market where model loyalty may be weaker than many providers would like. When platforms make it easier to compare and route between models, users can follow the best combination of price and capability.
For OpenAI and Anthropic, the issue is not only whether their systems remain ahead. It is whether the premium remains worth paying for a growing range of company workloads.
For Chinese AI companies, the opportunity is clear. If models from DeepSeek, Z.ai, and others continue to be viewed as competitive while staying far cheaper, they can keep gaining traffic from users who care about cost efficiency.
The key takeaway is simple: AI adoption is becoming more price-sensitive. As more companies move from experimentation to regular production use, the economics of tokens can matter as much as the model leaderboard.