Open-weight models process more than 70% of tokens and nearly one in four dollars in AI Gateway index

Open-weight models now process more than seven in ten tokens and account for nearly one in four dollars on Vercel AI Gateway, according to the October 2026 Production Index. Their share of spend has more than quadrupled since June, rising from one in twenty dollars to nearly one in four. This shift occurs within a rapidly expanding market where September gateway spend was 13 times December’s level, while the average price per token dropped 47%.

Spend Growth and Model Mix

Growth in open-weight usage is driven by both low-cost and high-cost models. DeepSeek-v4.1-Flash alone processed nearly half of all September tokens but represented just 2.6% of spend, with a price per token about a fifth of the open-weight average. Conversely, models priced at least four times the open-weight average, led by Kimi K3, drove about two-thirds of the increase in open-weight spend. Customers are increasing budgets on both sides of the market: frontier model spend grew tenfold since December, and 68% of teams that increased open-weight spending also increased closed-weight spending.

Provider Dynamics and Chinese Models

A significant portion of spend on Chinese-origin models is routed through US-based providers. In September, nearly 70% of gateway spend on Chinese models went through US-based providers, up from less than 10% in December. DeepSeek models processed 58% of all gateway tokens for 6.6% of spend, with two in five dollars spent on them going through US providers. Moonshot AI (creator of Kimi) accounted for 3.6% of tokens and 13.2% of spend, with US providers serving nine in ten of those tokens. Z.ai (creator of GLM) accounted for 7.2% of tokens and 4.5% of spend, with US providers serving seven in ten.

September Launch Scorecard

September saw notable launches with distinct adoption and revenue profiles:

• Jev: The fastest-adopted model in AI Gateway history, reaching more than a third of paid teams in its first two weeks. However, as a decision model, it accounted for just 0.15% of gateway spend and 1% of tokens.
• Opus 5.5: Captured more than one in ten gateway dollars in its first two weeks, the largest spend share of any September launch. This contributed to Anthropic’s overall share of gateway spend falling below 50% for the first time, dropping from 63.7% to 43.9%.
• GPT-6 Luna: Reached nearly three out of every twenty paid teams and had the highest retention rate of the month, with almost two-thirds of first-week users returning in week two.
• GPT-6 Astra: Reached a smaller audience but captured about one in twenty gateway dollars, more than five times the spend share of Luna.

Teams are increasingly treating models as replaceable components, with nearly one in three teams changing their most-used lab in September, an index record. High-volume teams (over 1 billion tokens) used an average of 42 different models between December and September.