AI Agent Cost Calculator
Estimate model-token charges per task from your task volume, workflow steps, model calls per step, and average tokens per call.
Choose a model
Set your usage
These are editable scenario inputs, not usage benchmarks.
Monthly model calls = active users × tasks per user × steps per task × LLM calls per step. Every call uses the selected model and the token averages below.
Use an average across calls, including the instructions and context sent to the model.
Include reasoning or thinking tokens if your provider bills them as output. This estimate does not infer them.
Review the estimate
Estimate for GPT-6.1 Sol at public standard text rates checked October 17, 2026.
Cost at different volumes
Each row keeps the same token assumptions.
| Active users | Model calls / month | Monthly | Annual |
|---|---|---|---|
| 100 | 300 | $0.9000 | $10.95 |
| 1.0K | 3.0K | $9.00 | $109.50 |
| 10.0K | 30.0K | $90.00 | $1,095.00 |
| 100.0K | 300.0K | $900.00 | $10,950.00 |
Monthly estimates use 30 days. Annual estimates use 365 days.
Estimate covers public standard text input and output rates only. If your provider bills reasoning or thinking tokens as output, include them in your output-token estimate; this calculator does not infer them. It excludes cached-token pricing, batch discounts, external tools and search, embeddings, vector stores, hosting and infrastructure, negotiated discounts, free-tier credits, and image, audio, video, or voice usage.
Save your results
Get your cost estimate sent to your inbox. We'll also send tips on how to reduce your AI spending.
No spam. Unsubscribe any time.
Budget the model calls in each task
A multi-step workflow can send several requests to a model before it completes a task. Set your own tasks per user, steps per task, and model calls per step. The estimate multiplies those inputs once, then applies the selected model’s token rates to each call.
Input and output fields are averages per model call. The calculator uses one selected model for every call, so treat the result as a model-token estimate rather than a full workflow budget. Add billed retries to your call assumptions. Price external search, tools, databases, and hosting separately.
