appConfigEntityId

bodyType

customContents

entityId

entityUuid

isVideo

articlePageQuery

article

articleSchemaContent

content

hooksArticlePageAdvertisementConfigArticle

hooksArticlePageFloatingButtonConfigArticle

appConfig

filter

json

liveArticle

Minxiao Chang

Xinmei Shen

The usage of open-weight AI models from China hit a record high on a popular US web development platform, driven largely by DeepSeek’s latest lightweight model, as business demand for Anthropic’s cutting-edge Fable 5 stalls due to high costs.
Open-weight models accounted for 54 per cent of token volume on Vercel’s AI Gateway on Tuesday, surpassing the 46 per cent consumed by proprietary models, according to data on its website.
Open models reached a record share of 62 per cent on the platform on Saturday, dwarfing closed systems’ 38 per cent, Vercel CEO Guillermo Rauch said on social media.
This was in stark contrast to the breakdown on June 24, when open-weight models accounted for only 28 per cent of Vercel AI Gateway’s token volume, compared with 72 per cent for closed ones, according to Rauch.
As of Tuesday, DeepSeek-V4-Flash was the most-used model in terms of token volume, while Chinese models dominated the top five, including Step 3.7 Flash from StepFun, GLM-5.2 from Zhipu, or Z.ai, and DeepSeek-V4-Flash’s updated 0731 version, which ranked second, fourth and fifth, respectively. The third spot was taken by OpenAI’s GPT-5.6 Luna, Vercel data showed.
The figures indicate that developers are increasingly choosing cheaper and open models for production workloads, particularly for autonomous agents that can consume large numbers of tokens for reasoning, writing code and calling software tools.

“Nowadays, open-source models are more widely used than closed models,” Tomasz Godziek, head of equities at Swiss bank J. Safra Sarasin, said in a media briefing in Hong Kong on Monday. “This is an important driver of the democratisation of AI.”
Godziek cited recent remarks by CloudFlare chief executive Matthew Prince that automated agents and bots now generated more than half of web requests for the first time. The proliferation of open-source AI models was a “very good enabler” of this trend, Godziek said.
However, their customizability has also raised security concerns. Taiwanese cybersecurity firm TeamT5 said hacking groups linked to Beijing were using models including DeepSeek to automate routine tasks and help develop malware, Bloomberg reported on Monday.
Vercel’s data reflects activity on its own platform rather than the broader AI market. However, separate corporate spending data points to growing resistance to the prices charged for the most advanced proprietary models.
Anthropic’s Fable 5 represented only 6 per cent of tokens that businesses purchased from the firm in the month after its release, according to an August report from US expense management company Ramp. Despite being Anthropic’s most expensive model, it generated only 11.4 per cent of spending on the company’s models.
By comparison, OpenAI’s GPT-5.6 Sol accounted for 25 per cent of OpenAI tokens and 23 per cent of spending. Fable 5 generated about three-quarters as much model-attributed spending as GPT-5.6 Sol.
Ramp said the results suggested businesses had found an upper limit to what they would pay for additional AI performance, particularly as open-weight alternatives narrowed the capability gap.
The shift has coincided with increased US investment in domestic open-weight models capable of competing with Chinese offerings.
Nvidia planned to use a US$6 billion licensing agreement with coding AI start-up Poolside to build a powerful open-weight model that could challenge DeepSeek and Moonshot AI’s Kimi K3, the Wall Street Journal reported over the weekend, citing people familiar with the matter.
The chip giant would separately invest US$1 billion in Poolside and hire more than 100 of its employees to work on Nvidia’s Nemotron open-model project, according to the report.
The deal is not a conventional acquisition, as Poolside’s founders will remain with the company.
Nvidia’s push highlights an emerging divide in the AI market: proprietary frontier models continue to capture substantial spending, but lower-cost open-weight systems are taking a growing share of the workloads that developers actually run.


Philippine tribe vows to stand ground against US-led Pax Silica: ‘listen to us’

As the US weighs restrictions on cloud computing, how will China’s AI sector adapt?

Chinese AI chipmaker Hygon plots expansion from data centres to robotics

For China’s hi-tech IPOs, it’s Hong Kong vs the mainland – but how much does it matter?

Artificial intelligence (AI)

DeepSeek leads surge in low-cost Chinese open-weight models on US platform

The DeepSeek app is seen on a smartphone screen in this arranged photo taken January 26, 2025. Photo: Shutterstock Images

People visit the Moonshot AI booth, featuring the Kimi K3 model, during the World Artificial Intelligence Conference in Shanghai. Photo: AFP

Alibaba’s lightweight Qwen model takes on larger AI systems from OpenAI, DeepSeek, Zhipu

Chinese tech titan Alibaba Group Holding’s new lightweight AI model Qwen3.8-27B has matched much larger rivals including OpenAI, DeepSeek and Zhipu. Photo: Shutterstock

OpenAI-backed legal tech firm pivots to Chinese Kimi K3 open-weight model

The Moonshot AI logo is displayed alongside its Kimi app on a mobile phone, in this illustration picture taken July 24, 2026. Photo: Reuters

There is a growing shift by Western tech firms towards Chinese open-weight systems amid soaring development costs. Photo: Getty Images

Tech

Tech Trends

News

China Future Tech

Open models reached a record share of 62 per cent on a US web development platform on Saturday, a reversal of figures in June


Open models reached a record share of 62 per cent on a US web development platform on Saturday, a reversal of figures in June.


DeepSeek leads surge in low-cost Chinese open-weight models on US platform

.css-1c6uqr6{color:inherit;font-weight:inherit;font-size:inherit;font-family:inherit;line-height:inherit;overflow-wrap:break-word;}Open models reached a record share of 62 per cent on a US web development platform on Saturday, a reversal of figures in June

Open models reached a record share of 62 per cent on a US web development platform on Saturday, a reversal of figures in June