AI Agent Cost Calculator

It depends almost entirely on how many model calls the loop makes. At this page's default workload: 500 runs a month, 5 calls per run, 4,000 input and 800 output tokens per call, which is 2,500 model calls in total. The same agent costs about $1.30 a month on GPT-5 nano and about $200.00 on Claude Fable 5. That spread is why model choice and call count matter more than token price shopping.

Five is a reasonable planning default for a tool-using agent, and it is what this calculator assumes. A single reasoning step plus a tool call and a read of the result is already three, and retries, reflection steps, and follow-up tool calls push it higher. Loops that plan, act, and check their own work commonly land between five and fifteen calls per run. Count the calls in one real trace before you budget, because this number multiplies everything else.

Managed Agents has no separate per-agent subscription: the model tokens bill at the same published per-million rates as a direct API call, so the rates in the table above apply. What changes is your visibility into the call count, because Anthropic runs the loop rather than your code. That makes the per-session budget the practical cost control, and it makes estimating calls per run more important, not less. Model the loop here first, then set the session budget to match.

Multiply your monthly agent runs by the LLM calls each run makes, then by the tokens per call: input tokens are billed at the model's input rate per million tokens and output tokens at its output rate. Agent loops surprise people because every reasoning step, tool call, and retry is its own model call. Five calls per run at 4,000 input tokens each is 20 million input tokens per thousand runs.

It depends on the tier, not the vendor. Every provider's fast tier costs a fraction of its frontier tier: on the same loop, GPT-5 nano runs about $1.30 a month while Claude Fable 5 runs about $200.00, and Claude Haiku 4.5, GPT-5.4 mini, and Gemini 3.7 Flash all land within a few dollars of each other. Output tokens are typically 3–8× input price everywhere. Enter your own workload above and the cheapest badge goes to whichever model genuinely wins for your numbers.

Usually, at real volume, because of structure rather than rates. An autonomous agent spends most of its model calls deciding what to do next; a Rills workflow encodes that routing as deterministic logic, which is free, and calls the model only for steps that genuinely need AI. Fewer model calls means a smaller bill, and every consequential step can wait for your approval at $0. At tiny volumes the base subscription can cost more than the raw token bill, and the calculator shows that honestly.

One AI credit covers $0.01 of underlying model cost. Each model call is rounded up to a whole credit with a one-credit minimum. Every plan includes a monthly credit pool; beyond it, overage is billed per 100 credits at your plan's published rate. Triggers, logic, and human approvals never consume credits, and a workflow paused for approval costs $0.

A chat question is one model call. An agent run is a loop: it reasons, calls tools, reads results, and reasons again, commonly five or more model calls per run, each carrying the conversation context as input tokens. Cost grows with calls per run, and with context size as the loop accumulates history.

Where a provider publishes a cached-read rate, input tokens served from cache cost roughly 10% of the full input price. Agent loops re-send a lot of identical context, so a high cache share can cut input cost substantially. Use the cache slider above to model it; models without a published cached rate ignore the slider.