Chatbot cost grows with conversation context
A chatbot request usually sends more than the latest user message. System instructions, previous turns, retrieved knowledge and tool definitions can be included again on every model call, causing input usage to grow as the conversation continues.
Estimate an early turn and a later turn separately. A realistic monthly budget should reflect the average number of model calls per conversation, not only registered users or opened chat sessions.
| Chatbot component | When it is charged | Optimization lever |
|---|---|---|
| System instructions | Usually sent with every request | Shorten or cache stable instructions |
| Conversation history | Grows across turns | Summarize or trim older messages |
| RAG context | Added when knowledge is retrieved | Return fewer, more relevant chunks |
| Assistant response | Generated on every model call | Set an appropriate output limit |
Translate message cost into chatbot unit economics
Cost per message is useful for engineering, while cost per 1,000 messages and cost per resolved conversation are easier to use for product pricing and capacity planning. Compare all models with the same context and response assumptions.
Retries, moderation, routing, agent handoffs and multi-model workflows can add requests that are invisible in a simple one-message estimate. Add those calls to the expected monthly interaction count or model them as separate scenarios.
| Business metric | Calculation | Best use |
|---|---|---|
| Cost per message | One request input plus output | Model and prompt comparison |
| Cost per conversation | Average request cost times turns | Support and sales planning |
| Cost per 1,000 messages | Per-message cost times 1,000 | Pricing plans and gross-margin estimates |
| Monthly chatbot cost | All model calls at monthly volume | Budgeting and model routing decisions |
Official Pricing Sources
Provider prices and billing rules can change. Verify the current rates and special pricing conditions before committing a production budget.