Open-Weight Models Now Take 71.5% of Gateway Tokens
Vercel's AI Gateway production index reports open-weight models now process over 71% of all gateway tokens and nearly a quarter of total spend, up more than fourfold since June. Nearly one in three teams switched their primary model provider in September, an index record, showing how volatile model choice has become for production apps.
Why it matters
๐ป Developer ยท If you're hardcoding a single model provider, this data says you're increasingly the outlier; routing flexibility is becoming the norm.
๐ฆ Product ยท Fast provider-switching among teams suggests model choice is now a tactical decision, not a permanent architecture commitment.
๐จ Design ยท Not directly design-facing, but shows the underlying AI infrastructure your product sits on is shifting fast and cheaply.
๐ Business ยท Falling reliance on closed-model vendors and rapid provider switching weakens vendor lock-in leverage, useful context for procurement negotiations.
๐ค Just Curious ยท Companies building AI products are increasingly choosing free, open versions of AI models instead of paying premium providers like OpenAI.