DeepSeek-V4-Flash ranked as the most-used AI model by token volume on Vercel's AI Gateway as of Tuesday, 25 August 2026, with Chinese open-weight models collectively dominating the platform's top five — a dramatic reversal that underscores growing developer preference for low-cost alternatives over expensive proprietary systems.

Record share for open-weight models

Open-weight models reached a record 62 per cent of token volume on Vercel AI Gateway on Saturday, with closed proprietary models accounting for the remaining 38 per cent, according to Vercel CEO Guillermo Rauch, who disclosed the figures on social media. By Tuesday, open-weight models still held 54 per cent of token volume versus 46 per cent for proprietary models, per data on Vercel's website.

The shift is stark when compared with 24 June, when open-weight models accounted for only 28 per cent of Vercel AI Gateway's token volume against 72 per cent for closed models, according to Rauch.

Chinese models dominate the top five

As of Tuesday, DeepSeek-V4-Flash led all models by token volume on the platform. Step 3.7 Flash from StepFun, GLM-5.2 from Zhipu (also known as Z.ai), and an updated DeepSeek-V4-Flash 0731 version ranked second, fourth, and fifth, respectively, according to Vercel data. The only non-Chinese model in the top five was OpenAI's GPT-5.6 Luna, which held the third position.

The breadth of Chinese representation — spanning DeepSeek, StepFun, and Zhipu — signals that the open-weight surge is not a single-model phenomenon but a structural shift across the China-based AI ecosystem.

Why it matters: Cost pressure reshapes developer choices

The figures indicate that developers are increasingly choosing cheaper, open models for production workloads, particularly for autonomous agents that consume large numbers of tokens for reasoning, writing code, and calling software tools. Business demand for Anthropic's cutting-edge Fable 5 has reportedly stalled due to high costs, according to reports, highlighting the price sensitivity driving this migration.

Token-intensive agentic workflows amplify cost differentials between proprietary and open-weight models, making even modest per-token price gaps significant at scale for enterprises and developers running continuous pipelines.

The competitive backdrop

Vercel's AI Gateway, a widely used US-based web development platform, serves as a real-time barometer of developer model preferences, making its token-share data a credible proxy for production-level adoption trends. The platform's data shows that within roughly two months, the open-weight share more than doubled — from 28 per cent on 24 June to a peak of 62 per cent by late August 2026.

As Chinese open-weight models continue to close the capability gap with frontier proprietary systems, the pressure on Anthropic, OpenAI, and other closed-model providers to justify premium pricing will only intensify in the months ahead.