Open models just hit 38% of enterprise AI
New data puts numbers on a shift everyone felt coming. Per AI.cc’s 2026 AI API Infrastructure Report — 2.4 billion API calls across 8,000+ accounts, via Open Source For You — enterprise AI token costs fell 67% year over year, from $18.40 to $6.07 per million, while open-weight models jumped from 11% to 38% of enterprise token volume. Microsoft, separately, rolled out its own models to cut its reliance on OpenAI.
The emerging pattern is a tiered stack: route the high-volume, everyday work to cheap open models and reserve a frontier model for the rare hard problem. Once most of your volume already runs on open weights, the obvious question is why rent them at all.
- The market is already voting for open models. Owning them — instead of renting per token — takes the marginal cost down to roughly electricity.
- “Cheaper per token” still bills per token. The only number that stops climbing is the one you own.
Sources: Open Source For You (AI.cc report); Microsoft via CNBC. See which model fits.