"I woke up to a $340 bill. One skill went into a loop overnight."
That's a real post from the OpenClaw Reddit community. The response thread had 47 comments, most of which were variations of: "same thing happened to me."
OpenClaw is free to download. Running it is not free. And the costs are remarkably easy to lose track of.
Why API Costs Are Hard to Predict
Unlike a SaaS subscription where you pay a fixed monthly fee, OpenClaw's costs are entirely dependent on usage. Every message processed, every skill invoked, every context window filled — it all adds up at your AI provider's per-token rate.
The problem isn't the pricing model itself. The problem is the opacity.
The three hidden cost multipliers
- Skills that loop. A poorly written skill can enter a retry loop and make hundreds of API calls before anyone notices.
- Context window explosion. Long conversations accumulate context. A 10-message thread might cost 50x more than a 1-message thread because the entire history is re-sent each time.
- No per-skill attribution. By default, OpenClaw doesn't tell you which skill is responsible for which spend. You see a total at the end of the month, not a breakdown.
The Real Numbers
One user in a Discord server I'm in documented their costs over 3 months. They were running a customer support bot for a small e-commerce store — not a high-volume operation. Their breakdown:
| Month | Total API Cost | Surprise? |
|---|---|---|
| Month 1 |