Agent loop cost calculator
See what an AI agent stuck in a loop costs per hour, overnight and over a weekend. Pick a model, enter how big each call is and how often it fires.
Input tokens include the system prompt, tools and conversation history sent on every call. In agent loops that history is usually most of the cost.
| One call | |
| Overnight (8 hours) | |
| A weekend (48 hours) | |
| 30 days nonstop |
Burn alerts are on Pro (7-day free trial). They check OpenAI, Anthropic, xAI and DeepSeek spend every hour and ping you by email, Slack or Discord.
How it's calculated
Cost per call = uncached input tokens × input price + cached input tokens × cached price + output tokens × output price, all per million tokens. Then multiplied by calls per minute and time.
Preset prices are standard list prices from the OpenAI and Anthropic pricing pages, checked November 13, 2026. Prices change, so edit any price to match your bill. Not included: cache write fees, batch or priority discounts, taxes, tool-call fees and data-residency surcharges. Reasoning or thinking tokens are billed as output, so count them in output tokens.
Why loops get expensive
An agent re-sends its whole context on every step. A loop that retries every few seconds, or a long session that keeps growing its history, can multiply a normal day's spend in a few hours. Provider dashboards often show the spend hours later, and monthly budgets only warn you once a lot is already gone.
How to catch one early
- Cap steps and retries in your agent code.
- Use prompt caching for the parts of the context that don't change.
- Set an hourly spend alert, so a loop shows up the same day instead of on your invoice. AI Spend Tracker does this across providers without a proxy or code changes.