Chinese AI Models and Products Gain Structural Ground on US Rivals

Chinese AI models reached 61% of weekly token consumption on OpenRouter by mid-2026, up from a minority share through most of 2025, with American AI startups now routing significantly more application traffic to Chinese LLMs.[1][2] Tencent launched WorkBuddy in late May 2026, reported as China's #1 PC productivity AI agent by daily active users shortly after launch, and extended it globally via Tencent Cloud.[5][4] OpenAI's 2024 API cutoff for China restructured the developer ecosystem, with Chinese model providers absorbing the displaced user base.[8][9] The overall shape: Chinese AI products have moved from domestic catch-up into competitive positions at both the API/model layer and the enterprise productivity application layer.

The OpenRouter data is a practical proxy for developer routing behavior, not just domestic Chinese adoption—meaning U.S.-based startups are actively choosing Chinese models. If the shift reflects capability parity rather than pure price competition, it suggests Chinese AI providers can compete for developer infrastructure spend globally, not only within China.

Through most of 2025, U.S. models led weekly token consumption on OpenRouter, the API routing platform widely used by developers and AI startups. From early 2026, Chinese models became the primary growth driver, and by mid-2026 they accounted for 61% of the platform's weekly token market share.[1][2] American AI startups are now routing significantly more application traffic to Chinese LLMs than they were a year prior.[1] OpenRouter data is not a comprehensive view of the global LLM market, but it is a practical proxy for developer routing decisions—making the share figure a concrete data point on where developers are directing traffic, regardless of where they are headquartered.

Observers differ on what the OpenRouter trend means. One reading, from developer-community commentary, is that the LLM market now operates like an inference spot market: developers route to whichever model delivers the best results at the lowest cost at any given moment, with no lasting commitment.[3] On this view, Chinese models' gains are real but contingent on price—a gap U.S. labs could close. A different reading, offered by analysts tracking the data, treats the growth as evidence of a more durable competitive shift in which Chinese models have achieved capability sufficient to capture organic developer preference.[1]

On the product side, Tencent launched WorkBuddy in late May 2026 as a PC-based productivity AI agent, initially for China and then for global users via Tencent Cloud.[4] The product reads files, calls tools, writes reports, builds slide decks, analyzes data, and assumes over 100 predefined expert roles.[5] It integrates with third-party platforms including GitHub, Jira, Notion, Gmail, Google Drive, and Slack—a configuration directly competitive with enterprise productivity tools built on U.S. AI platforms.[5] Tencent subsequently released an enterprise edition, framing the move as scaling from individual AI augmentation to team-level AI workflows.[6] By early June 2026, multiple sources described WorkBuddy as China's #1 PC productivity AI agent by daily active users.[5][4] Analyst Jeffrey Towson characterized Tencent's strategy as a deliberate push from single-user agents toward full enterprise suite products.[7]

Part of the structural context is supply-side: OpenAI terminated API services for developers in China in mid-2024, removing a major external dependency from the Chinese AI developer ecosystem.[8][9] Chinese model providers—including Baidu, Alibaba, and Moonshot AI—moved quickly to recruit the displaced developer base.[8][9] That displacement accelerated domestic Chinese model adoption and likely contributed to the competitive trajectory visible in OpenRouter data by 2026, though the direct causal link is not fully established in available reporting.