Plan AI costs around real workloads
Explore common AI scenarios, understand what drives their cost, and open the appropriate calculator when one is available.
Available use cases
Choose the scenario closest to your workload.
Chatbot / customer support
A support or sales chatbot that holds a back-and-forth conversation with a user. Cost is driven mainly by how many turns a conversation runs and how much history gets re-sent on every message.
What drives the cost- Conversations per month and messages per conversation
- Average message length (user and assistant)
- System prompt size, repeated on every request
- How much prior conversation history is kept in context
Meeting summary
Summarizing a meeting transcript into notes and action items. This covers summarization only — converting audio to text (transcription) is a separate service with its own, separate cost, and is not included.
What drives the cost- Meetings per month and average duration
- Transcript density (how much text a minute of speech produces)
- Summary detail and optional action-item extraction
Document summarization
Summarizing a report or other long text document you already have. Cost is driven by document length and the level of summary detail requested — not by page count, which has no single defensible token conversion.
What drives the cost- Documents per month and document length (measured in words)
- Summary detail and optional key-points extraction
- Scanned or image-based documents need OCR first, which is not covered here
Document Q&A
Answering questions using document context that has already been retrieved (retrieval-augmented generation, or "RAG"). This estimates only the model-inference step after retrieval — not the retrieval infrastructure itself.
What drives the cost- Questions per month and average question length
- How much retrieved context is supplied per question
- Embeddings, vector-database, and retrieval infrastructure cost are not included
Not sure which fits? View all calculators →
Planned use cases
Each calculator needs its own verified token assumptions and pricing checks before it can produce an honest estimate — we'd rather ship a calculator that's right than one that guesses. These scenarios are shown so you can see what cost factors they'll cover once built; they don't have a working calculator yet, and this page does not estimate their cost. Read the methodology to see how every calculator's numbers are sourced and verified.
Internal AI assistant
A single assistant used by a team internally for search, drafting, and Q&A over internal knowledge combined — cost scales with headcount and how often each person uses it. If retrieval-based Q&A is the main cost driver you care about, the document Q&A calculator already estimates that slice on its own.
- Number of active users and requests per user per day
- A mix of drafting, search, and retrieval-based Q&A in one tool
- System prompt and instruction length
Coding / app generation
Generating or modifying code with a large repository or file as context. Large context sizes approach the context-window limits of many models.
- Size of the code or repository context sent per request
- Output length (generated code is usually verbose)
- How often the same large context is re-sent versus cached
Why AI costs differ by use case
Every scenario above reduces to the same handful of levers — they just apply to different units of work:
- Volume. How many conversations, meetings, documents, or questions happen in a month.
- Input size. How much text goes into the model each time — a message, a transcript, a document, a retrieved context chunk.
- Output size. How much the model generates back — output is usually billed at a higher rate than input.
- Model selection. Which model you choose, since price per token varies significantly by model.
- Repetition. Content resent on every call, like a system prompt or accumulating conversation history, multiplies its cost across every request.
For the full calculation logic behind these levers, see the methodology.