Chinese AI models are 60–90% cheaper — route your workloads

The frontier AI you've been paying premium prices for has quiet competition, and a lot of American companies just noticed. As token prices at the top US labs keep climbing, businesses are quietly rerouting the boring 90% of their AI work to Chinese open-weight models that are "good enough" — and a fraction of the cost. If you've wired your product or your ops to a single lab's API, this is the moment to ask a cheaper question: why are you paying frontier prices for work a commodity model can do?

What actually happened

On July 7, 2026, CNBC reported that Chinese open-source models now cost 60% to 90% less than comparable models from Anthropic and OpenAI, per OpenRouter's Justin Summerville. That gap is moving real money. AI startup Lindy shifted all of its traffic from Anthropic's Claude to DeepSeek; CEO Flo Crivello said the switch would "save the company millions of dollars within months." His blunter version, to Rest of World: "You don't need God to write your email."

The performance gap is narrower than the price gap suggests. Z.ai's GLM 5.2 scored within a percentage point of Anthropic's Opus 4.8 on a closely watched agentic benchmark — at roughly one-fifth the cost, per the same CNBC reporting. The behavior is showing up in the aggregate numbers: Chinese models have taken more than 30% of the tokens processed on OpenRouter every week since February 8, 2026, peaking near 46% — up from an 11% average over the prior year. GLM 5.2's daily token volume on Vercel grew about 27x in its first full week after launch.

Why it matters for your business

Here's the trap most teams fall into: you build against one lab's SDK, hard-code its model name, and now your cost structure is whatever that lab decides to charge. When prices climb — and they have — you eat it, because switching means a rewrite.

The companies saving millions aren't loyal to a flag or a lab. They route. Simple summarization, classification, drafting, and tagging go to a cheap model. Hard reasoning, agentic multi-step work, and anything customer-facing stays on a frontier model until a cheaper one proves it can hold the line. The abstraction that makes this possible — a routing layer you own, where the model behind each task is a config value, not a load-bearing dependency — is the whole game. Build it once and every future price war works for you instead of against you.

Sources: CNBC, International Business Times, Rest of World.