Practical guide
AI Agent Cost per Run: A Step-by-Step Estimate
Estimate an AI agent run from model-call count, tokens per call, and tool fees, including retries and branching scenarios.
Updated · Sources checked
Map one complete run
An agent run can invoke the model repeatedly: plan, call a tool, inspect the result, revise, and answer. Count every model call in that sequence. The AI Agent Cost Calculator prices GPT-5 input and output tokens across that call count and adds an optional tool cost to every call.
For a research agent that averages six model calls, 3,000 input tokens and 600 output tokens per call, enter 6, 3000, and 600. If only two calls use a paid search tool, do not enter the full search price as the per-call tool cost. Spread it across all six calls: two $0.01 searches equal $0.02 per run, or about $0.003333 per model call. The calculator then produces a modeled run cost and a 12-run reference.
Turn traces into realistic averages
Group successful production traces by task type. Record model calls, total input tokens, total output tokens, and paid tool charges for each trace. Use a median case for the baseline and a high percentile for the budget ceiling. Failed runs matter too: retries and abandoned branches still consume tokens.
For monthly planning, multiply the displayed run cost by expected runs yourself. The “at 12 runs” figure is only a convenient multiple; the calculator does not know your monthly volume.
The estimate excludes caching, model-specific reasoning meters, storage, search fees unless entered, taxes, and discounts. It also applies one average token count and one average tool charge to every call, so irregular workflows require a weighted average or separate scenarios.
FAQ
Is “agent steps” the same as user requests?
No. One user request may trigger several model calls. Use the calls visible in an execution trace.
How should I model retries?
Either include retry calls in the observed average or run a second scenario with a higher call count. Do not count only completed paths.