About

What is this?

I started How Much To Run AI to answer a practical question: for a given workload and budget, does it make more sense to run a model ourselves or use a commercial API?

Token costs can add up, and powerful models can feel slow when compute is limited. Meanwhile, open-weight models and the software and infrastructure for running them keep improving. Self-hosting is now a practical option, but its cost and speed depend on the model, workload, hardware, and usage.

I built a cost calculator and compare open-weight and closed models, as well as self-hosting and commercial APIs. I also collect published deployment experiences to help you weigh those results against your own workload and budget.

Use the calculator, model comparisons, and published deployment experiences to see which option fits your tasks and budget. As models and the tools for running them improve, the trade-offs keep changing.

Why the thinking face?

The logo uses the thinking face emoji (🤔) — because that's exactly the expression on your face when your boss asks "so how much does it cost?"

Logo artwork adapted from Noto Emoji, © Google Inc., licensed under Apache 2.0.

Contact

For questions, suggestions, or feedback, email us at hello@howmuchtorunai.com or DM @YeeJan_neutral on X.