AI chatbot pricing comparison (2026): what platforms actually cost
How AI chatbot pricing really works in 2026 — per seat, per message, per resolution, and usage models — and how to model your true cost before buying.
TL;DR: AI chatbot pricing comes in four main shapes — per seat, per message/credit, per resolution, and usage-based — and the headline number rarely reflects your real cost. Model it on your actual volume, watch for what counts as a billable event, and compare total cost of ownership including the human team and add-ons.
The hardest part of buying an AI chatbot is not comparing features; it is comparing prices, because vendors bill on different units. A "cheap" per-message plan can cost more than an "expensive" flat plan at your volume, and a per-resolution model shifts risk in ways a per-seat model does not.
This guide explains the main pricing models, what to watch for in each, and how to estimate your true cost. Specific vendor numbers change constantly, so this focuses on the models rather than quoting figures that will be stale next quarter — always confirm current pricing on the vendor's site.
The four main pricing models
| Model | You pay for | Best when | Watch for |
|---|---|---|---|
| Per seat | Each human agent | Human-heavy support | AI value not tied to cost |
| Per message / credit | Each message or AI credit | Low, predictable volume | Cost spikes at scale |
| Per resolution | Each resolved conversation | High resolution rate | Definition of "resolved" |
| Usage-based | Consumption (tokens, data) | Custom/embedded builds | Estimating consumption |
Per-seat pricing
You pay per human agent, often with AI features bundled or added per tier. Simple and predictable, but the cost is tied to your team size, not to how much value the AI delivers. Common in help-desk platforms.
Watch for: whether AI features are included or a paid add-on, and whether you pay for seats that mostly supervise an AI that does the work.
Per-message or per-credit pricing
You pay per message or per "AI credit," sometimes bundled into monthly allowances. Attractive at low volume, but costs can climb fast as traffic grows, and "credit" definitions vary — some models cost more credits than others.
Watch for: what consumes a credit, how overage is billed, and how quickly cost scales with a traffic spike.
Per-resolution pricing
You pay when the AI resolves a conversation. This aligns cost with outcomes, which many teams like, but it makes two things critical:
- The definition of "resolved." Confirm exactly what counts as a billable resolution — and what happens with partial or incorrect resolutions.
- Volume sensitivity. At high resolution volume, per-resolution can exceed flat plans. Model it.
See the Intercom Fin AI review for a resolution-priced example.
Usage-based pricing
Common for custom or embedded builds: you pay for consumption — model tokens, data processed, or API calls. Flexible and fair, but you need to estimate consumption, which depends on prompt size, retrieval context, and traffic.
Watch for: how to forecast consumption, and whether costs are observable in real time so a runaway prompt doesn't surprise you.
How to model your true cost
- Estimate volume — conversations or messages per month, with a growth and a spike scenario.
- Map to the billing unit — translate volume into seats, messages, resolutions, or consumption.
- Add the human team — most deployments still need human agents; include them.
- Add-ons and tiers — the features you need may require a higher tier.
- Compare total cost of ownership, not the entry price, across your realistic scenarios.
The cheapest headline price frequently loses once you plug in real volume and the features you actually need.
Hidden costs to check
- Overage rates when you exceed an allowance.
- Feature gating that pushes you to a higher tier.
- Onboarding or setup fees.
- Integration costs if connectors are paid or custom.
- The cost of wrong answers — a cheap bot that misinforms customers has a real cost that never appears on the invoice.
How Currai fits
For usage-based and custom builds, the cost you actually incur is model and retrieval consumption — and it is invisible without instrumentation. Currai tracks token use and cost per conversation, so you can see what each answer costs, catch a runaway prompt, and set budgets and alerts. See track token cost and budgets and alerts for LLM cost.
Frequently asked questions
What is the most common AI chatbot pricing model?
Per-seat and per-message/credit models are common in packaged platforms; per-resolution is growing for AI agents; usage-based is typical for custom and embedded builds. Each shifts cost and risk differently.
Is per-resolution pricing cheaper?
It depends on your volume and resolution rate. It aligns cost with outcomes, but at high resolution volume it can exceed flat plans. Model it against your real numbers, and confirm what counts as a billable resolution.
How do I estimate my chatbot cost?
Estimate conversation volume with growth and spike scenarios, translate it into the vendor's billing unit, add your human team and any required tiers or add-ons, and compare total cost of ownership rather than entry price.
What hidden costs should I watch for?
Overage rates, feature gating to higher tiers, setup fees, paid integrations, and the unbilled cost of wrong answers. Confirm current pricing and terms on the vendor's site.