Vercel's October Index Says the Model Is Now a Replaceable Part

Vercel's October AI Gateway Production Index has a headline figure: open-weight models now process 71.5 percent of gateway tokens and 23 percent of spend, and their share of spend has more than quadrupled since June.

The number we would frame on the wall is a different one. In September almost one in three teams changed their most-used lab, an index record. Teams processing at least a billion tokens used an average of 42 different models. The report's own conclusion is that the model is becoming a replaceable part of a much larger investment.

This is Vercel's data from its own gateway, so it reflects Vercel's customers, not the whole market. Still, the pattern fits what we see building client tools: the prompt, the evaluation set and the glue code are the durable assets, and the model name is a config value.

The practical advice is dull and useful. Keep a small test set of real tasks from your own work. Route through a layer that lets you swap models in one line. Then re-run the set whenever a cheaper option appears, because with prices per token falling 47 percent since December, the saving is often there for the taking.