[Digital Today reporter Jinju Hong (홍진주)] Open-weight models released by Chinese artificial intelligence (AI) labs have surpassed U.S. closed AI models in developer usage. As competition moves beyond performance and cost efficiency emerges as a key criterion in choosing AI models, Chinese models are rapidly expanding their influence, an analysis shows.

Cryptopolitan, a blockchain media outlet, reported on July 18 that on OpenRouter, a platform that routes developer API calls across multiple AI models, weekly token throughput for Chinese AI models was more than three times that of U.S. models.

The backdrop is price competitiveness. OpenAI and Anthropic sell access to closed models on a per-token basis, while Chinese AI labs such as DeepSeek, Moonshot AI and Z.ai are releasing open-weight models that developers can download and use directly. The analysis says developers are starting to treat operating costs as an important selection criterion alongside performance when building or optimizing AI systems.

Actual usage also reflects the shift. As of late June, the six most-used models on OpenRouter were all open models from Chinese companies. They included Tencent, Xiaomi, DeepSeek, MiniMax and Z.ai.

This is the first time Chinese model usage has surpassed U.S. models since February this year. From Feb. 9 to 15, Chinese models processed 4.12 trillion tokens on OpenRouter, while U.S. models logged 2.94 trillion tokens. That marks a major change in the market structure compared with most of 2025, when U.S. models accounted for about 70 percent of top usage.

The price gap is also clear. Cryptopolitan said Z.ai's GLM 5.2 charges $1.40 per 1 million input tokens and $4.40 per 1 million output tokens. Anthropic's Opus 4.8, by contrast, charges $5 and $25, respectively. It says developers are shifting to Chinese open-weight models as cost gaps widen even among model groups that offer similar performance.

The trend is also seen on other platforms. Chinese open models accounted for 41 percent of downloads on Hugging Face this spring, surpassing U.S. models. In the Production Index published by Vercel, the share of open-weight models expanded to 29 percent in June from 11 percent in April. These models handled about one-third of total token throughput, but related costs did not reach 4 percent of total spending.

DeepSeek accounted for 22.6 percent of token throughput on the Vercel AI Gateway, ranking as the second most-used model group after Google.

Companies' burden of AI operating costs is also fueling the spread of open-weight models. In a KPMG survey of 2,145 corporate executives, 29 percent of respondents said they did not properly understand or control their companies' AI operating costs.

Uber, for example, used up its AI coding budget for the year by April and then limited each engineer's monthly AI tool usage to $1,500. It also introduced a case in which an unnamed company was billed $500 million for Anthropic Claude usage in a single month.

Separately from cost competitiveness, security and regulatory issues remain unresolved tasks. For example, criticism has emerged that Z.ai's cloud API falls under China's National Intelligence Law, raising the possibility that Chinese authorities could access code and data sent through that route. The U.S. Congress also launched an investigation in May into security risks that could arise from using Chinese AI models in critical infrastructure.

Chinese companies plan to further strengthen their open-weight strategy. Moonshot AI is preparing to launch an open-source model, 'Kimi K3', with 2.8 trillion parameters. The company claimed the model can compete with top-tier U.S. AI systems.

Kimi K3 posted slightly lower scores than Anthropic's Claude Fable 5 and OpenAI's GPT-5.6 Sol in composite benchmarks, but it was found to outperform Claude Opus 4.8 and GPT-5.5 in coding and AI agent evaluations.

Pricing also emphasized competitiveness. Kimi K3 was set at $3 per 1 million input tokens and $15 per 1 million output tokens. That is cheaper than GPT-5.6 Sol at $5 and $30, and Claude Fable 5 at about $10 and $50.

The industry expects Chinese open-weight models to gain more influence by emphasizing cost efficiency and deployment flexibility as performance gaps gradually narrow. The outlook says companies will need to consider not only cost savings but also data control, security and country-by-country regulatory risks.