AI Token Cost Calculator
Estimate your AI API costs before you spend. Compare GPT, Claude, and Gemini pricing in seconds.
$2.5/1M input tokens · $10/1M output tokens · Context: 128,000 tokens
Used to project your monthly cost (assumes 30 days).
Input Cost
$0.0025
Output Cost
$0.0050
Total Cost / Request
$0.0075
Projected Monthly Cost
Based on your requests/day input
$22.50
Compare All Models
Cost for the same 1,000 input / 500 output tokens, cheapest first.
| Model | Provider | Input Cost | Output Cost | Total Cost | Context Window |
|---|---|---|---|---|---|
| Gemini 1.5 FlashCheapest | $0.0001 | $0.0002 | $0.0002 | 1,000,000 | |
| GPT-4o mini | OpenAI | $0.0002 | $0.0003 | $0.0005 | 128,000 |
| Claude Haiku | Anthropic | $0.0008 | $0.0020 | $0.0028 | 200,000 |
| Gemini 1.5 Pro | $0.0013 | $0.0025 | $0.0038 | 1,000,000 | |
| GPT-4o | OpenAI | $0.0025 | $0.0050 | $0.0075 | 128,000 |
| Claude Sonnet | Anthropic | $0.0030 | $0.0075 | $0.0105 | 200,000 |
| Claude Opus | Anthropic | $0.0150 | $0.0375 | $0.0525 | 200,000 |
AI Model Cost Calculator
Most cost estimates assume one user action equals one model call — but agents, RAG pipelines, and multi-step chains rarely work that way. A single "ask the AI" action might trigger a planning call, a tool call, a retrieval step, and a final generation call, each consuming tokens on its own. Growfiy's calculator estimates the cost of an individual call; this page is about summing those calls into a real per-feature or per-task cost.
Built for teams building agentic features, RAG-based search or support tools, or any workflow where the AI part of a user action is actually several calls stitched together.
How Multi-Step Workflows Multiply Cost
- Agent loops: planning, tool selection, tool execution feedback, and final response can each be a separate call — a 3-4 step agent task can cost 3-4x a single-call estimate before any retries.
- RAG pipelines: an embedding call for retrieval plus a generation call that includes the retrieved passages as extra input tokens — larger or more numerous retrieved chunks directly raise the generation step's input cost.
- Chains and self-correction: a pipeline that checks its own output and retries on failure adds a partial extra call's worth of cost, proportional to how often that retry path triggers.
- Conversation memory in agents: if prior steps' outputs get resent as context in later steps, token cost within a single task can grow with each additional step, not stay flat.
Modeling Cost for a Complete Feature
Start by listing every model call a single completed task actually makes, in order — not the happy-path minimum, but the typical path including any tool calls or validation steps. Estimate the input and output tokens for each step separately using the calculator above, since later steps often carry more input tokens than earlier ones once prior context gets included.
Sum those per-step estimates into one per-task cost, then add a margin for your expected retry rate. That per-task number — not the cost of any single call in isolation — is the one worth using for budgeting, pricing a feature, or deciding whether a workflow is economically viable at scale.
Frequently Asked Questions
Why does a single feature cost more than one model call suggests?
Many AI features aren't a single request — an agent workflow might plan, call a tool, reason over the result, and generate a final answer, each as a separate model call. A RAG pipeline often adds a retrieval-and-rerank step plus the generation step. Estimating from one call undercounts the real cost significantly.
How do I estimate the cost of an AI agent that takes multiple steps?
Map out the typical number of steps a task takes end to end — planning, tool calls, intermediate reasoning, final response — estimate the token cost of each step separately using the calculator above, then sum them for a per-task total rather than a per-call total.
Do retries and self-correction steps add meaningfully to cost?
Yes, and they're easy to undercount. If an agent or pipeline validates its own output and retries on failure, even a modest retry rate (say, 10-20% of tasks needing one extra pass) can add a proportional amount to your average per-task cost across the board.
How does RAG affect the total cost per user query?
A typical RAG query involves an embedding call for the search, a retrieval step, and then a generation call that includes the retrieved context as additional input tokens — larger retrieved context means more input tokens on the final generation step, on top of the base query cost.
Should I estimate cost per feature or cost per model call?
Cost per feature (or per completed task) is the more useful number for budgeting and pricing decisions, since it reflects what a user action actually costs end to end — cost per model call alone hides how many calls a single feature triggers.
How do I keep multi-step workflow costs from spiraling?
Cap the maximum number of steps or retries a workflow can take, use a smaller/cheaper model for intermediate reasoning or tool-selection steps and reserve the flagship model for the final response, and cache repeated context (like tool schemas or system instructions) across steps in the same run.