Agent Cost Calculator
SourcesStart free

Interactive planning tool

AI agent cost calculator

Build a monthly estimate that includes requests, model tokens, cache hits, retries, tool calls, RAG, observability and human review.

Calculate your AI agent cost

Start with a working assumption and adjust the inputs to your own workload.

Workload
Tasks completed without needing a rerun
Model usage
Source-linked public list pricing; update before planning spend
Agent behavior
Extra requests after failures or incomplete outputs
Traces, logs, evaluation and retention allowance
Optional escalation or quality-review allowance
Visible uncertainty around your chosen assumptions
RAG & tools
Third-party APIs, browser or search services
Embeddings, vector reads and storage allowance

Estimated monthly cost

$0

Planning range: —

↘ Model your drivers
$0.000Cost / request
$0.000Cost / successful task
—Cost / 100 users

Cost breakdown (monthly)

Model tokens$0
Tools & RAG$0
Observability$0
Human review$0
Retries$0
Top driver: model tokens

Try caching stable context to reduce repeated input cost.

⌁ How the estimate works

Monthly cost combines model tokens, tools, RAG, observability, human review and retry behavior from your editable assumptions.

monthly = model + tools + RAG + observability + human review + retries

▣ Sources & assumptions

Rate records are source-linked and versioned. Always check the provider page before final budgeting.

OpenAI pricing ↗Gemini pricing ↗Methodology →

? FAQ

Why can an invoice differ? Contracts, taxes, regional pricing, tool metering and usage behavior can change the final bill.

Reconcile an invoice →

What the calculator includes

Each request can represent a single completion or a multi-step agent task. The calculator multiplies per-step model usage by the number of requests, then adds retry load and the per-task operational allowances you enter for tools, retrieval, observability and human review. The visible low-to-high range only applies the uncertainty percentage you choose; it does not claim to predict provider billing changes.

The tool is deliberately explicit about its limits. It cannot infer your negotiated rate, discounts, taxes, provider-specific rounding, regional capacity fees, image/audio metering or the number of individual queries performed by a search tool. Use it for planning, then reconcile it with exported usage after launch.

Frequently asked questions

Does API list pricing equal my monthly cost?

No. List pricing is one input. Your bill can also reflect output length, cache behavior, retries, tool calls, storage, regional or contract terms, taxes and provider-specific metering.

Why estimate cost per successful task?

Requests that fail, retry, or need manual recovery still consume resources. Cost per successful task connects technical usage to the outcome your product actually delivers.

Can I use custom pricing?

Yes. Select Custom rates in the calculator and enter your own input and output price per million tokens. This is useful for negotiated or provider-specific terms.