Automation Planning · Free Tool
AI Agent Cost Calculator
Estimate what an AI agent or LLM workflow costs to run per task, per month, and per year across Claude, GPT, and Gemini models, including multi-step calls and prompt caching.
Same workload on every model
| Model | Per task | Per month |
|---|
Standard-tier list prices for short prompts, checked October 2026 (Gemini Flash prices are introductory until 31 December 2026). Excludes cache-write surcharges, batch discounts, long-context pricing, and tool fees such as web search. Edit the price fields to model anything else.
About this tool
Most LLM cost calculators price a single prompt. Real agents make several model calls per task (plan, call a tool, read the result, decide, answer), and each call re-sends the growing context. This calculator models that: set the tasks per month, model calls per task, and average input and output tokens per call, and it works out cost per task, per month, and per year.
Prompt caching matters a lot for agents, because the system prompt and tool definitions repeat on every call. Set the share of input tokens served from cache and the calculator applies each provider's cheaper cache-read rate. A comparison table then prices the same workload on every listed model, so you can see what moving routine steps to a smaller model would save.
Common uses
- Budget the running cost of an AI agent before building it
- Compare a frontier model against a cheaper one for high-volume steps like classification or extraction
- See how much prompt caching would cut an existing LLM bill
Frequently asked questions
How many tokens is a typical agent call?
A short system prompt plus a few tool definitions is often 2,000 to 5,000 tokens before the user input, and context grows with every step as tool results are added. Outputs are usually much smaller, a few hundred tokens for a tool call or short answer. If you have a working prototype, use its real usage numbers instead.
What is prompt caching and why does it matter?
Providers can cache the unchanging start of a prompt (system prompt, tool definitions, reference documents) and bill later reads of it at a fraction of the normal input price, often 5 to 10%. For an agent that re-sends the same instructions on every call, this can cut input cost by more than half.
Are the model prices current?
They were taken from the providers' published pricing in October 2026 (standard tier, short-context prompts). Prices change and some are introductory, so every price field is editable. Batch processing, long-context surcharges, and fine-tuning are not included.
Is my data uploaded to a server?
No. This tool runs entirely in your browser using JavaScript, so nothing you type is transmitted, logged, or stored anywhere. You can even use it offline once the page has loaded.