The running cost of an AI assistant for a small business depends mostly on the model underneath. Google's Gemini 4 Hydrogen, the cheapest model in the Gemini 4 family, still costs 15 times more per token than Claude Haiku 5.5, so a support bot or document pipeline should be priced across vendors, not inside one vendor's lineup.
As Simon Willison breaks down, Hydrogen costs $1.50 per million input tokens and $7.50 per million output. Its bigger siblings, Carbon and Argon, cost $2/$10 today and double to $4/$20 on December 1. Haiku 5.5 and GPT-6 Luna sit at $0.10/$0.50. Hydrogen also becomes Google's free tier model.
What this means for a support bot or invoice pipeline
Most small-business tasks don't need a frontier model. Sorting leads, pulling fields from invoices and answering FAQs fall into that group. I usually route the work: a cheap model handles the bulk, and an expensive one only gets the hard cases. That split is a big part of what drives cost in my projects. The December doubling is a second lesson. Keep the model as a config setting, so a price change means editing one line. One caveat: Simon's tests showed a stark quality gap between Hydrogen and its siblings, despite the close price.
Why it matters: Run 50 real tickets or invoices through two or three models and multiply the result by your monthly volume. That number is your budget, whatever the vendor calls "cheap".
Source: Simon Willison