Vercel Says Open-Weight Models Hit 29% of AI Gateway Volume
Vercel says open-weight models accounted for 29% of AI Gateway token volume in June, up from 11% in April, according to the company's July Production Index. The report is based on anonymized aggregate routing data from Vercel's AI Gateway, which routes model requests from production applications across AI providers.
The shift did not lower the overall average token price in June. Vercel said gateway token volume rose 29% and spend rose 27%, leaving the average price per token roughly flat after a nearly 20% increase in May. The company attributed the balance to cheaper open-weight traffic rising at the same time that closed-weight frontier model prices increased by about 12% per token.
DeepSeek was the largest open-weight contributor in the report, reaching 22.6% of gateway tokens and moving within two percentage points of Google's token share. Vercel also highlighted Z.ai's GLM 5.2, saying its daily token volume grew about 50x between its June 16 API availability and the end of the month.
The index suggests enterprises are using open models for a growing share of production workloads, but not replacing frontier providers wholesale. Vercel said the top four U.S. frontier labs still captured 95% of AI Gateway spend in June, with Anthropic alone accounting for 61% of spend on 32% of tokens.