Chatbot Cost Calculator
How much will your customer support chatbot cost at scale? Enter your parameters and get a full cost breakdown.
How this calculator works
A support chatbot's bill has more moving parts than a single API call. This calculator estimates output tokens from the response length you enter, then estimates input tokens automatically as a system prompt plus conversation history plus the user's message — since input size scales with how much context each reply carries. It multiplies those token counts by your chosen model's price, using the current prices shown in the calculator, across your monthly conversation volume.
On top of inference, it adds a hosting cost tier sized to your conversation volume and peak concurrency, an optional vector database cost if you enable retrieval-augmented answers, and an optional monitoring/observability cost. All four add up to a total monthly cost, which the calculator also expresses per conversation and compares against a benchmark cost for human support agents so you can see the relative savings.
Monthly cost breakdown
inputTokens ≈ 500 (system prompt) + 2×outputTokens (history) + 100 (user message)
inference = (totalInputTokens ÷ 1,000,000)×inputPrice + (totalOutputTokens ÷ 1,000,000)×outputPrice
total = inference + hosting + vectorDB + monitoring
How to use it
- 1.Enter your expected monthly conversation volume.
- 2.Set the average tokens per response — longer, more detailed answers cost more per conversation.
- 3.Pick the model your chatbot uses.
- 4.Choose a hosting provider (or "none" if you already have infrastructure) and your peak concurrent users.
- 5.Toggle on a vector database and/or monitoring if your chatbot uses retrieval or observability tooling, then compare the total against the human-support benchmark.
Common mistakes
- ⚠Underestimating conversation history — multi-turn conversations carry prior messages as input tokens on every turn — a chatbot that references earlier context costs more per conversation than a single-turn estimate suggests.
- ⚠Forgetting infrastructure costs — inference is often the smallest line item once you add hosting, a vector database, and monitoring — model all four, not just the model price.
- ⚠Sizing hosting for average, not peak — peak concurrent users, not average daily volume, determines whether your hosting tier can keep up during busy periods.
- ⚠Assuming free tiers scale — a vector database or monitoring free tier that works at 1,000 conversations a month can require an upgrade well before 100,000 — recheck provider limits as you grow.
Frequently asked questions
What drives the biggest share of chatbot cost?▾
It depends on volume. At low conversation counts, inference (the model calls) usually dominates. At high volume, hosting, a vector database, and monitoring can catch up to or exceed inference cost — the calculator's breakdown shows the split for your numbers.
How are input tokens estimated if I only enter response length?▾
The calculator assumes roughly a 500-token system prompt, conversation history worth about twice the response length, and a 100-token user message. If your real prompt or history is longer or shorter, the input-token estimate will shift accordingly — treat it as a starting point, not an exact count.
Is an AI chatbot actually cheaper than human support?▾
For most support use cases, yes — the calculator compares your total monthly cost against a benchmark human-agent cost per conversation so you can see the order of magnitude for your own volume and model choice, using the current prices shown in the calculator.
Do I need a vector database for a support chatbot?▾
Only if you want the bot to answer from your own knowledge base (help docs, past tickets) rather than general knowledge. If your bot only handles simple scripted flows, you can leave it off and save that cost entirely.
What infrastructure does the hosting estimate assume?▾
A managed compute tier (AWS, Azure, GCP, or Railway) sized to your conversation volume and scaled slightly for peak concurrency. If you're running on existing infrastructure with no incremental cost, choose "none" for hosting.
How do I reduce chatbot costs shown here?▾
Try a cheaper model in the dropdown, shorten your system prompt (it drives input tokens on every turn), cap response length, or route only complex queries to a pricier model while handling routine ones with a cheaper one.