Full Breakdown
Surge of Chinese Open-Weight AI Models on Vercel’s Platform
8/25/2026, 8:42:56 PM
Core Event
On a recent Tuesday, open-weight AI models from China accounted for 54 % of token volume on Vercel’s AI Gateway, overtaking proprietary models, which fell to 46 %. This marked a sharp rise from the 28 % share recorded on June 24, when closed models dominated with 72 % of tokens. Vercel’s data show that DeepSeek-V4-Flash was the most-used model by token count, with Chinese offerings occupying four of the top five positions.
Background & Context
Open-weight models—publicly available AI systems that can be freely integrated—have become attractive to developers seeking cost-effective alternatives to commercial offerings such as Anthropic’s Fable 5. Rising business demand for lower-priced solutions has driven the shift toward Chinese models, which are positioned as lightweight and inexpensive.
Data & Statistics
- Open-weight models: 54 % of token volume on Tuesday vs. 28 % on June 24.
- Proprietary models: 46 % on Tuesday vs. 72 % on June 24.
- Top models by token volume: DeepSeek-V4-Flash (1st), DeepSeek-V4-Flash 0731 (2nd), Step 3.7 Flash (3rd), GLM-5.2 (4th), Z.ai (5th).
- OpenAI’s GPT-5.6 Luna placed sixth, indicating a modest presence among the leading models.
Official Statements & Responses
Vercel CEO Guillermo Rauch highlighted the record share of open-weight models on social media, noting the platform’s AI Gateway now sees a majority of traffic from these models. He contrasted the current figures with the June 24 baseline, emphasizing the rapid adoption trend.
Why It Matters
The dominance of Chinese open-weight models signals a growing market preference for affordable, high-throughput AI solutions, potentially reshaping the competitive landscape for U.S. and European providers. As developers prioritize cost and performance, the surge may encourage further investment in open-weight model development and influence future pricing strategies for proprietary AI services.
