Chatbot Cost Calculator
Models real conversations — growing history, system prompt, caching — not naive per-message math. Prices updated 2026-08-02.
- Per conversation
- $—
- Tokens per conversation
- —
- Caching saves
- $—/mo
- Cheapest capable tier
- $—/mo
Assumes history grows each turn and is resent (how chat APIs actually work).
Why chatbot costs surprise people
Naive estimates multiply "cost per message" by message count — and land 2–3× off. Real chat APIs are stateless: every turn resends the system prompt plus the whole conversation so far. By turn 8, you're paying for the first message for the eighth time. That compounding is why the calculator simulates the conversation turn by turn instead of using flat averages.
The compounding also means prompt caching is not optional for chatbots at scale: the resent history is exactly what caching discounts by ~90%. On an 8-turn conversation, roughly 80% of all input tokens are cacheable.
Frequently asked questions
› How much does it cost to run an AI chatbot per month?
For most small products: $20–$400/month. A chatbot handling 500 conversations a day on a mini-tier model costs roughly $20–60/month, while the same volume on a flagship model runs $300–900/month. The two levers that matter are model choice (10–30× difference) and prompt caching (up to ~2× savings on busy bots).
› Why does conversation history make chatbots expensive?
Every message you send to the API must include the conversation so far, so input tokens grow with each turn. A 10-turn conversation can resend the same history 10 times. This is exactly what prompt caching fixes: the resent history is billed at ~10% of the normal input rate.
› What is a typical conversation size in tokens?
A support-style exchange of 6–10 turns typically totals 3,000–8,000 input tokens (system prompt + growing history, resent each turn) and 800–2,000 output tokens. Our calculator models this growth for you instead of assuming a flat per-message cost.
› Should I use GPT, Claude, or Gemini for my chatbot?
For routine support conversations, the mini/flash/haiku tiers of any major vendor perform well and cost 10–30× less than flagships. A common production pattern is routing: the cheap model handles everything, and the ~10% of hard conversations escalate to a flagship. The comparison table shows your exact workload priced on every model.