WSWhat Scene?

Learn · Pricing & costs

What does an AI feature or tool cost to build?

Two costs: building the feature and running it. The build (designing it, connecting your data, testing for accuracy) is usually the larger, one-time figure, comparable to a web app project. Running it is the ongoing AI usage, billed per token, which is often surprisingly small for typical volumes.

As of July 2026, mid-tier AI models cost roughly $2 to $3 per million input tokens and $10 to $15 per million output tokens, so a normal question and answer costs a fraction of a cent. The running bill only becomes significant at high volume or with very large inputs on every call.

The part that determines success isn't the model fee. It's the engineering around it: giving the model the right data, testing for accuracy, and building guardrails so it fails safely. That's where the real budget goes, and where cheap builds cut corners.

Updated July 2026

Want this built, not just explained?