DeepSeek leads surge in low cost Chinese open-weight models on US platform
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,…
A record surge in low-cost Chinese open-weight models is dominating the U.S. web development platform Vercel's AI Gateway, as business demand for premium models like Anthropic's Fable 5 slows due to high costs. Open-weight models made up 54% of token volume on Tuesday, surpassing proprietary models at 46%. This marks a significant shift from June, when open models only accounted for 28% of the token volume, with proprietary models making up 72%.
Chinese models are leading the pack, with DeepSeek's lightweight model DeepSeek-V4-Flash being the most-used model in terms of token volume. Other Chinese models include Step 3.7 Flash, GLM-5.2, and Z.ai, all of which ranked in the top five. OpenAI's GPT-5.6 Luna rounded out the top five.
This trend towards cheaper and open models is driven by businesses looking for more affordable production workloads, especially for autonomous agents that consume large numbers of tokens for reasoning, writing code, and calling software tools. Tomasz Godziek, head of equities at Swiss bank J. Safra Sarasin, noted that open-source models are now more widely used than closed models, contributing to AI democratization.
However, the customizability of these models has raised security concerns, with Taiwanese cybersecurity firm TeamT5 reporting that hacking groups linked to Beijing are using models like DeepSeek to automate tasks and develop malware.
Vercel's data reflects activity on its own platform, but corporate spending data indicates a growing resistance to the prices of advanced proprietary models. Anthropic's Fable 5, the company's most expensive model, only accounted for 6% of tokens purchased by businesses in the month after its release, generating only 11.4% of spending.
In contrast, OpenAI's GPT-5.6 Sol accounted for 25% of OpenAI tokens and 23% of spending. The shift towards open-weight alternatives has narrowed the capability gap, leading businesses to find an upper limit to what they're willing to pay for additional AI performance.
Written by urgent.news from South China Morning Post's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
This story
This is one outlet's version. Read the fullest account.
