How Much Does an AI Chatbot Cost in 2026? A Developer Pricing Guide

"How much will an AI chatbot cost?" is one of the most common pre-build questions — and one of the most mis-answered. List prices on provider sites describe per-token rates, not what a chat product spends in production. This guide gives developers a worksheet you can reuse in planning decks, fundraising models, and architecture reviews.

The variables that actually matter

1. Monthly active users who open the bot

2. Sessions per user and turns per session

3. Input tokens per turn — user text + system prompt + conversation history + RAG context

4. Output tokens per turn — average assistant length

5. Retries and tool calls — each retry is another billed request

Ignore (1) at your peril: a "simple" bot with a 2,000-token system prompt and ten-turn history grows faster than spreadsheet guesses.

Quick estimation formula

Let:

• U = monthly active users
• S = sessions per user per month
• T = turns per session
• I = average input tokens per turn (fully loaded)
• O = average output tokens per turn
• R = price per million tokens (blended input/output for a rough estimate)

Monthly tokens ≈ U × S × T × (I + O)

Monthly cost ≈ (Monthly tokens / 1,000,000) × R

Add 25–40% headroom for retries, eval traffic, and internal testing unless you hard-cap usage.

Three worked scenarios

Support widget (small B2B SaaS)

• 800 MAU, 2 sessions, 5 turns
• I = 1,200 (history + short RAG), O = 250
• Tokens/month ≈ 800 × 2 × 5 × 1,450 ≈ 11.6M tokens

At a blended $3/M tokens → ~$35/month. At premium model rates (~$15/M) → ~$174/month.

Copilot inside a dev tool

• 3,000 MAU, 8 sessions, 6 turns
• I = 2,500 (code context), O = 400
• Tokens/month ≈ 3,000 × 8 × 6 × 2,900 ≈ 417M tokens

At $3/M → ~$1,251/month. This is where token invoices surprise teams who prototyped on tiny prompts.

Internal HR policy bot

• 200 employees, 4 sessions, 3 turns
• I = 900, O = 200
• Tokens/month ≈ 200 × 4 × 3 × 1,100 ≈ 2.6M tokens — looks cheap until you add document RAG and logging.

For internal tools, also read Building AI-powered internal tools.

Hidden multipliers

• System prompts duplicated every turn
• RAG chunks appended every turn
• Long outputs when users ask for essays or code dumps
• Failed requests you still pay for on some providers

Our deep dive: The true cost of AI APIs.

Flat-rate comparison

Daymora charges $25/month flat for API access to GPT-5, Claude Sonnet 4, and Gemini 2.5 Pro — no per-token meter. Flat-rate wins when:

• Loaded token math already exceeds ~$20–$25/month or will next quarter at your growth rate
• You want to ship features without throttling context windows for cost

Flat-rate loses when usage is sporadic and tiny (weekend hacks, demos). See What is a flat-rate AI API?.

Build vs buy for the chat UI

This article covers model/API spend, not headcount. A custom React chat (see React streaming chatbot) is often a few dev-days; ongoing cost is API + observability + support.

Checklist before you commit

1. Log real I and O from staging with production-shaped prompts.

2. Multiply by expected U in six months, not today.

3. Compare token estimate to $25/month flat.

4. Read OpenAI vs Anthropic vs Gemini vs flat-rate if you are still shopping providers.

Summary

AI chatbot cost is a function of fully loaded tokens × traffic, not the size of the user's last message. Run the formula with honest inputs; add headroom; then choose billing shape (metered vs flat) that matches whether you are experimenting or shipping a core feature.