US developers turn to cheaper Chinese AI
American developers are routing real work to Chinese models like DeepSeek and Kimi, often at a tenth the price of Claude or GPT, to cut their AI bills.
American developers are quietly moving real work onto Chinese AI models to cut costs. Rest of World reported on June 17 that operators including a San Diego developer and the founder of AI assistant company Lindy now pay for Chinese models such as DeepSeek, Moonshot’s Kimi, Qwen and Minimax alongside their US subscriptions. The pull is price: one developer runs coding tasks on DeepSeek for under $0.50 an hour against roughly $10 an hour on Claude, and DeepSeek’s published API rates sit near $0.14 per million input tokens and $0.28 per million output, many times below frontier US models.
The shift shows up in platform data. Rest of World cites Vercel figures showing DeepSeek’s share of tokens jumping from under 1% to 17% in May, and OpenRouter’s most-used list now leans on DeepSeek, Tencent, Minimax and Xiaomi.
What it means for operators
You do not need a frontier model for every job. As Lindy’s Flo Crivello put it, “You don’t need God to write your email.” Routine drafting, classification and extraction can run on a cheaper open model while you keep a pricier one for the hard cases. Open-weights models keep closing the gap, and the difference is wide enough to show up on a monthly bill.
Two cautions. Test any model on your own tasks before you trust it, the way you would when choosing any AI tool. And know where your prompts go: a hosted Chinese API sends your data to that provider, which matters if you handle customer records. Fold it into your stack on purpose, knowing what each provider gets to see.