Cook with AI Enrollment closed
AI & LLMs Runs locally

Inference Cost Calculator

Estimate your monthly LLM inference bill and compare models across OpenAI, Anthropic, Google, and more.

← All tools No signup · Nothing uploaded
ELI5

It estimates what your AI feature will cost each month.

A little more detail

What this tool does

An AI request can include input tokens, output tokens, cached tokens, and many calls per user. This calculator combines those numbers so you can compare a realistic monthly bill across models.

Use it to
  • Budget a chat, generation, or agent feature before launch
  • Compare models using the same traffic and token assumptions
  • See which input, output, or usage change saves the most money
Interactive workspace Results update as you type
What are you building?

Daily active users
Usage assumptions

(all editable)

Agent steps count as calls, not runs.

System prompt + history + retrieved context. ~4 chars ≈ 1 token.

Output tokens cost 4–6× more than input on most models.

Share of input tokens billed at the provider's cached rate (~90% cheaper).

publishes no cached-input discount, so the cache rate is ignored for it.

Estimated monthly inference bill ·

/mo

per user / month

per LLM call

Fresh input Cached input Output

calls · input tokens · output tokens per month

Suggested for these assumptions
at scale
Daily active users Monthly bill Per user
Every model, same assumptions

Sorted by monthly bill. Click a row to select that model.

Model Provider Tier $/mo $/user vs selected

Everything runs in your browser. A share link only encodes your inputs in the URL; no usage data is sent to a server.

This calculator turns your product assumptions into a monthly inference bill: daily active users, LLM calls per user, input and output tokens per call, and optional prompt-cache hit rate. Pick a workload preset (chat, RAG, agent, and others) to get sensible defaults, then tweak every slider.

Use it when you are planning an AI feature or comparing models before you wire up billing. The table ranks every model on the same assumptions so you can see cheap vs capable trade-offs, not just list prices.

Figures come from published API rates (29 models, July 2026). They ignore batch pricing, enterprise discounts, image or tool surcharges, and your own caching layer — treat the output as a planning estimate, not a vendor quote.