Agent Run Cost

Agent run cost is the unit cost of one completed agent execution after planning calls, tool-call context, retries, retrieval, cache, and infrastructure assumptions are included.

Question

How do I calculate cost per AI agent run?

Quick answer

Formula: refresh_share_usd = refresh_monthly_usd / monthly_agent_runs

Formula: cost_per_agent_run = generation + retrieval + reranking + refresh_share + vector_db + cache + infra

Formula: margin_pct = (run_price_or_chargeback - cost_per_agent_run) / run_price_or_chargeback * 100

Assumption: one completed run can contain multiple model requests.
Assumption: tool outputs and retrieved context are measured as prompt-side tokens.
Assumption: fixed refresh work is amortized across monthly agent runs.

Example: if generation, retrieval, reranking, refresh share, vector, cache, and infra sum to $0.18, one completed agent run costs $0.18 before any human approval or tool-use fee is added.

Scenario Table

Scenario	What changes	Expected effect
More tool steps	Model requests per completed run increase	Cost rises even if each individual model call is cheap
Heavier context	Prompt tokens include more tool output, traces, retrieved chunks, or state	Input-side spend and cache sensitivity move faster
Better cache reuse	Repeated runs reuse prior context or response blocks	Per-run cost drops without changing model quality

Where This Fits

Use Agent Run Cost when the decision unit is one completed agent execution. Use Coding Agent Cost per Task when accepted coding output and rework are the denominator. Use Break-even Price when you already know the blended user-month cost and need the required price.

Run the Agent Run Cost Tool

Model per-run cost from model calls, context, retrieval, cache, infra, and refresh assumptions.

Open Agent Run Cost