Chinese AI Models Like DeepSeek and GLM 5.2 Are Undercutting Claude and GPT on Price

Chinese AI models are no longer a curiosity for cost-conscious engineering teams in the US. According to a CNBC report published July 7, 2026, Chinese AI models like DeepSeek and Z.ai's GLM 5.2 are winning over American companies as Claude and GPT-5.6 costs keep climbing, and the shift is showing up in hard usage numbers, not just anecdotes.

The cost gap driving the shift

The core number is stark: open source Chinese models are running 60% to 90% cheaper than the leading Anthropic and OpenAI models, per the report. That is not a rounding-error discount. For a team burning tens of millions of tokens a month, a 90% price cut on comparable output is the difference between a five-figure and a six-figure monthly bill. CNBC frames this as a broader mood change among engineering leaders, who have moved from "adopt AI regardless of model" to actively price-shopping every workload.

Lindy's move from Claude to DeepSeek

The clearest illustration is AI startup Lindy, which moved 100% of its production traffic from Anthropic's Claude models to DeepSeek. Lindy's CEO described watching the cost curve "crash to the ground" after the switch, and expects the move to save the company millions of dollars within months. That is a full migration, not a partial hedge, and it is the kind of decision that gets easier every time a Chinese lab closes the quality gap another notch.

How big is Chinese AI's US market share now

The trend shows up clearly in OpenRouter data. The share of tokens routed by US companies to Chinese AI models has held above 30% every week since February 8, 2026, and has spiked as high as 46% in that window. Compare that to the twelve months before, when Chinese models averaged just 11% of routed traffic. That is roughly a threefold to fourfold jump in a matter of months, and it lines up with a string of competitive Chinese releases, including Z.ai's GLM 5.2 in June, which landed within a percentage point of Anthropic's Opus 4.8 on a closely watched agentic benchmark while costing around one-fifth as much, according to Vercel's adoption tracking.

The catch: it's not only about the sticker price

None of this means Claude or GPT are being abandoned wholesale. Enterprises with strict data-residency, compliance, or IP requirements are generally more cautious about routing sensitive workloads through Chinese-hosted infrastructure, and benchmark parity on one agentic test does not guarantee parity across every task type, tool-use pattern, or safety requirement a production system cares about. What has changed is that "cheaper" is no longer automatically read as "worse enough to ignore." Teams are increasingly running side-by-side comparisons on their own workloads rather than defaulting to whichever US lab shipped the flashiest demo. That same instinct, testing several models on the same prompt before committing spend to one, is the premise behind bring-your-own-key tools like ByteChat, which let you hold API keys for both US and Chinese providers and see the cost difference on your own traffic rather than a vendor's benchmark slide.

Frequently asked questions

Are Chinese AI models really cheaper than Claude and GPT?

Yes, according to the July 7, 2026 CNBC report, open source Chinese models are running 60% to 90% cheaper than leading Anthropic and OpenAI models, driven partly by lower inference costs and partly by aggressive pricing to win US customers.

Did a real company switch from Claude to DeepSeek?

Yes. AI startup Lindy moved 100% of its traffic from Anthropic's Claude models to DeepSeek, with its CEO stating the change would save the company millions of dollars within months.

How much of US AI traffic now goes to Chinese models?

On OpenRouter, the share of tokens from US companies routed to Chinese AI models has stayed above 30% every week since February 8, 2026, and reached as high as 46%, compared to an 11% average over the prior year.

Price gaps this wide tend to close fast in either direction, so today's cost comparison is worth rechecking in a few months, not treating as permanent.