OpenClaw's heartbeat runs a full agent turn every 30 minutes by default, whether or not you ask it anything. Pick your model and settings to see what that idle time costs a month, from each provider's published prices.
30m is OpenClaw's default (1h when Anthropic OAuth is set up).
500 is our assumption for a short check-in reply; change it to match your logs.
$0 while idle. Hearth has no heartbeat: when you aren't asking it anything, it makes no model calls at all. Everyday work runs on a small model on your own computer, and cloud calls only happen when you send or allow one, under a monthly cap you set that's checked before every call.
Runs a month = (24 h × 60 ÷ interval) × 30 days. Cost = runs × (input tokens × input price + output tokens × output price) ÷ 1,000,000.
The big number assumes no prompt caching. The range under it is the best case where every run's input is read from the provider's cache at the cached price. Real bills sit somewhere between, depending on whether the cache is still warm when the next heartbeat fires (many caches expire after 5 minutes; a heartbeat comes every 30).
Context per run: OpenClaw's docs say isolatedSession: true cuts a heartbeat from about 100K tokens of conversation history to about 2-5K. "Full history" uses 100K; "Isolated + light" uses 3.5K, the middle of that range. Your own sessions may be bigger or smaller: type your real numbers in.
This prices only the idle heartbeat. Anything you actually ask your agent to do costs extra on top.
| Model | Input | Cached input | Output | Source |
|---|
Prices are USD per 1M tokens, standard tier, checked 3 Oct 2026. DeepSeek's off-peak rate (outside 01-04 and 06-10 UTC on weekdays) is half the peak rate shown.
OpenClaw heartbeat: docs.openclaw.ai/gateway/heartbeat (default every 30m; "Heartbeats run full agent turns. Shorter intervals burn more tokens.")
Our free OpenClaw self-check reads your own config and installed skills and tells you if the gateway is reachable from your network, what your heartbeat is set to, and whether any skill matches the publicly reported malicious ones. It runs only on your machine and changes nothing.