AI Agent Cost Simulator

Estimate the monthly API cost of running an AI agent. Configure orchestrator and worker models, turns, tool calls, and see how context growth impacts your bill. Free, no sign-up.

Models

60%
40% orchestrator, 60% worker

Agent Behavior

5
2

Scale

Tasks/month: 1,500
LLM calls/month: 7,500
Tokens/task (in): 28.0K
Tokens/task (out): 1.5K
$0.059Per Task
$2.93Daily
$87.98Monthly
7,500API Calls/Month

Cost Breakdown

93%
7%
Orchestrator (Claude Sonnet 4.6): $0.054/taskWorker (Claude Haiku 4.5): $0.0043/task

Context Growth Per Turn

Each turn accumulates prior outputs and tool results, growing the input context.

Turn 1 (W)
3.0K
Turn 2 (W)
4.3K
Turn 3 (W)
5.6K
Turn 4 (O)
6.9K
Turn 5 (O)
8.2K
Optimization Tips
  • Enable prompt caching to save ~74% per task ($65.51/mo).

What If You Switched Orchestrator?

ModelProviderMonthly Costvs Current
Claude Sonnet 4.6 (current)Anthropic$87.98
GPT-5 NanoOpenAI$8.02Save 91%
Mistral Small 3.1Mistral$9.06Save 90%
Gemini 2.0 FlashGoogle$9.15Save 90%
Llama 3.3 70BMeta$9.51Save 89%
DeepSeek-V3DeepSeek$9.95Save 89%
Llama 4 ScoutMeta$10.37Save 88%
Llama 4 MaverickMeta$12.27Save 86%
DeepSeek-V3.2DeepSeek$13.24Save 85%

Save your results — get the cheatsheet

Drop your email and I'll send over my Agent Engineering Patterns (free PDF). You'll also join the newsletter — unsubscribe anytime.

Or preview the Agent Engineering Patterns landing page →

Estimates model a multi-turn agent where context grows each turn as prior messages accumulate. Actual costs depend on your tokenizer, context management strategy, tool output sizes, and provider pricing. Does not include infrastructure, vector DB, or rate-limit costs. All calculations run locally in your browser.

What the AI agent cost simulator does

An AI agent isn't one API call — it's a loop of them: an orchestrator model planning, worker models doing steps, tool calls, and a context that grows with every turn. This simulator lets you configure that shape (orchestrator + worker models, turns per task, tool calls) and estimates the monthly bill, including the cost that compounds as context accumulates. Free, no sign-up.

How to estimate agent cost

  1. Pick your orchestrator and worker models.
  2. Set turns per task, tool calls, and task volume.
  3. Read the monthly estimate and see how context growth changes it.

Why agents cost more than you expect

The trap is context growth. Because each turn usually re-sends the running conversation and tool outputs, input tokens climb turn over turn, so a 10-turn task can cost far more than 10× a single call. That's why the cheapest single model isn't always the cheapest agent — and why trimming context, capping turns, and routing simple steps to a smaller model matter. To price the underlying models, use the LLM pricing calculator.

Frequently asked questions

Why are AI agents expensive to run?

Context re-sent on every turn makes input tokens grow super-linearly, so multi-turn tasks cost far more than the per-call price suggests.

How do I cut agent costs?

Cap turns, summarize or trim context between steps, route simple sub-tasks to a cheaper worker model, and cache repeated tool results.

Is it free?

Yes — a free browser tool, no signup.

SharePost

More tools like this