Chinese AI Models Gain U.S. Users on Cost, Performance Advances

Chinese artificial intelligence systems are attracting a growing base of American users, driven by competitive pricing and improving capabilities that challenge U.S.-developed frontier models.

Raffi Krikorian, chief technology officer at Mozilla, switched to Moonshot's Kimi K3 model within days of its launch in July for daily tasks including calendar management and document processing. He previously relied on Z.ai's GLM-5.2 for similar work, finding both Chinese systems adequate replacements for more expensive American alternatives like Anthropic's Claude Fable.

"It just seems snappier," Krikorian told the Associated Press, which first reported these details.

Why it matters

The adoption of Chinese AI models by U.S. technology leaders and companies like Coinbase signals a fundamental shift in the global AI landscape. As these systems approach parity with American frontier models at a fraction of the cost, they threaten U.S. dominance in a strategic technology sector—even as Washington implements export controls designed to maintain that lead. The trend also highlights how price sensitivity, particularly for agentic AI applications that compound token costs, may override concerns about geopolitical competition.

Cost advantage drives adoption

Chinese models offer dramatic pricing advantages over American competitors. Technology executive Curt Meinhold, who founded digital legacy platform LilyList, noted that Chinese systems cost "a handful of cents per million output tokens versus 30 bucks or 40 bucks or 50 bucks" for comparable U.S. models.

This cost differential becomes especially significant for agentic AI applications, where systems autonomously execute complex, multistep tasks that multiply token usage. Goldman Sachs identified Chinese models as reaching a "critical stage" for widespread adoption, particularly as demand grows for these cost-intensive use cases.

OpenRouter data from the past month showed the five most popular AI models on its platform were all Chinese. Sensor Tower estimates indicated Kimi recorded more than 930,000 downloads in the week following K3's release—a 200% increase—with U.S. downloads jumping 387% to approximately 86,000.

Capability gaps remain

While Chinese models are becoming serious competitors, they still lag American AI leaders in overall capabilities, according to Anastasios Angelopoulos, co-founder and CEO of Arena, an AI evaluation platform.

Recent Chinese releases include Z.ai's GLM-5.2 in June, Moonshot's K3 in July, Alibaba's Qwen3.8 Max preview in July, and DeepSeek's V4 model preview in April. Experts assess these systems as nearly matching the intelligence of frontier models from OpenAI and Anthropic, though not surpassing them across all dimensions.

Open-source strategy and policy implications

Most Chinese AI models operate as open-source systems, contrasting with closed-source approaches from companies like Anthropic and OpenAI. This strategy allows broader examination and development, potentially accelerating global adoption.

U.S. policy decisions have occasionally created openings for Chinese competitors. Z.ai released GLM-5.2 shortly after the Trump administration imposed export controls on Anthropic's Fable and Mythos models, keeping them offline for over two weeks. "Restricting an American model can immediately create an opening for a Chinese competitor," Angelopoulos said.

The Trump administration accused Moonshot on Wednesday of using "covert" methods to build K3 from Anthropic's technology, though it stopped short of calling the practices illegal. Some U.S. politicians and companies have alleged Chinese startups engage in illicit "distillation" of American models—claims Beijing rejects.

Despite these tensions, Chinese AI startups face sustainability questions similar to their U.S. counterparts. Z.ai reported revenue of $107 million last year alongside a net loss of $694 million.

These details were first reported by the Associated Press.