Open-weight models hit 62% of Vercel tokens; Chinese LLMs lead research
Open-weight models have crossed to majority share of developer traffic on Vercel's AI infrastructure platform and lead academic research citations over American open models. The data notes that enterprise adoption of open-weight models is still early, with tooling still needing to adapt.
Open-weight models account for 62% of Vercel AI Gateway token volume as of August 2026, up from 28.4% in June, while closed models fell from 71.6% to 38% of share. In academic research, Chinese open LLMs appear in roughly 40% of recent AI/ML papers on arXiv, surpassing American open models at 25-30%, a reversal from 2024 when the ratio was about 10% Chinese versus 30% American. Qwen appears in about one in three papers that mention any LLM. Separately, SemiAnalysis tracked open-source model convergence across three successive eras of AI development and found that open-source models closed the gap with closed models in half the time of the prior era with each successive era.
Open-weight models reached 62% of Vercel AI Gateway token volume in August 2026, up from 28.4% in June, and Nathan Lambert's analysis of 500,000 arXiv AI/ML papers found Chinese open LLMs in roughly 40% of recent papers, up from about 10% in 2024, with Qwen in one in three LLM-mentioning papers.
SemiAnalysis found open-source models closed the gap with closed models in half the prior era's time across early scaling, reasoning, and agentic phases of AI development.
SemiAnalysis: open-source models closed gap with closed models in half the time each successive AI development era
Nathan Lambert's analysis shows Chinese open LLMs in 40% of AI papers; Qwen in one-third of LLM-mentioning papers
Open-weight models at 62% of Vercel AI Gateway tokens in August, up from 28.4% in June; August 22 noted as record day
I send a short email each morning with the stories that moved. If you would rather just read here, that works too.
Subscribe free