How Much Does an AI Chatbot Cost to Run? A Monthly Breakdown (2026)

Short answer: an AI chatbot has four cost layers: the platform plan, the AI model usage, WhatsApp or other channel fees, and your team's time. Platform plans range from about $14 to several hundred dollars a month. Model usage for a typical conversation is a fraction of a cent to a few cents, so 1,000 conversations can cost anywhere from about $2 to $180 in model fees.

Last updated: October 17, 2026. Written by the BossReply Team. BossReply is our product and is used as one example; all prices were read in October 2026 and can change, so confirm on each vendor's page.

Key takeaways

• The model is usually the smallest cost. On our worked example (6 messages each way, about 9,000 input and 1,800 output tokens), a cheap-tier model costs roughly $0.002 to $0.007 per conversation, and flagship models roughly $0.03 to $0.18.

• The platform's pricing model matters more than the AI. Per-contact, per-conversation, per-seat and flat plans give very different bills at the same volume.

• On WhatsApp, Meta charges separately for template messages. Replies inside the 24-hour window were free on Meta's page when we read it, but vendor notices say that may change.

• Bringing your own API key lets you pay the model provider directly at their listed price.

What are the cost layers of an AI chatbot?

An AI chatbot bill is the sum of four layers. Most pricing comparisons only show the first one.

How much does the AI model cost per conversation?

Model providers charge per million tokens, split into input (what the model reads) and output (what it writes). A token is roughly three-quarters of a word in English, so one million tokens is on the order of 750,000 words.

To compare models, we used one illustrative conversation: six customer turns, about 1,500 input tokens per turn (your instructions plus retrieved knowledge plus the question) and 300 output tokens per turn. That gives 9,000 input and 1,800 output tokens. The calculation is input price times 0.009 plus output price times 0.0018, using the per-million prices on each provider's page.

Prices come from the providers' official pricing pages, read in October 2026 (OpenAI's page is developers.openai.com/api/docs/pricing; Google's introductory rates for some models run only to the end of 2026). The arithmetic is ours. It assumes no caching, no reasoning-token overhead and no growing history; real conversations carry earlier turns forward, so inputs will often be higher.

The spread is about 100 times between cheapest and most expensive. For FAQ-style support where the answer sits in the retrieved text, the cheap tier is often enough, which is our opinion rather than a measured result. Test models on your own questions before deciding. Our comparison of OpenAI, Claude, Gemini and Grok for support chatbots goes deeper.

What does that look like per month?

Multiply the per-conversation cost by your volume. Using the figures above:

Two things can lower the real bill. Prompt caching: Anthropic lists cache reads for Haiku 5.5 at $0.01 per million input tokens against $0.10 for uncached input, which suits a fixed system prompt. And Anthropic's batch API gives a 50 percent discount, but it processes asynchronously, so it does not suit live chat.

How do platform plans compare?

Platform pricing models fall into a few types. The same chatbot traffic can cost very different amounts depending on which one you pick. For the full comparison read per contact vs per conversation vs flat pricing.

• Flat plans. BossReply costs $14 (Starter), $29 (Pro) and $69 (Agency) per month, with yearly billing saving up to 17 percent. Plans are not priced by stored contacts. Starter includes 100 live chat conversations and 500 bot-driven conversations per month, extra bot-driven conversations cost $1 per 100 per month, and Pro and Agency include unlimited conversations. See bossreply.com/pricing.

• Per active contact. ManyChat bills by contacts; one third-party review reported Essential at $17 for 250 contacts and Pro at $39 for 2,500, but ManyChat's own page blocked us and other sites quote different numbers, so check manychat.com. respond.io also counts monthly active contacts; its Growth plan was listed at $159 per month for 1,000, with overage of $12 per 100, and its structured data conflicted with the visible price.

• Per conversation. Botpress lists Plus at $150 per month with 250 conversations and extra packs at $0.65 per conversation, plus AI spend included at $0.10 per conversation.

• Credits or replies. Stammer.ai lists a Standard plan at $49 per month with 500 AI replies and $0.01 per extra message. FastBots uses credits, with extra 1,000 credits at $10.

• Per seat. WATI includes 3 to 5 users depending on plan and charges for extra users on higher plans; its base prices were not visible to us.

Whatever you compare, calculate the bill at your real volume. A per-contact plan that looks cheap at 250 contacts can change sharply at 5,000, while a flat plan stays predictable.

How does bring-your-own-key change the cost?

With bring-your-own-key (BYOK), you create an account with the model provider and the platform sends requests with your key, so you pay the provider's listed token price directly. BossReply lets you choose the model (OpenAI, Claude, Gemini or Grok) and bring your own API key. The upside is price transparency and the ability to switch models as prices move. The downside is a second bill and the need to watch your usage and set spending limits. For a longer explanation see AI chatbot platforms with bring-your-own-key.

What do WhatsApp and other channels add?

On WhatsApp, Meta charges per delivered template message since July 1, 2025, with rates by category and recipient country. Marketing templates are always charged. Utility templates are charged outside an open service window and free inside it. Meta's page, as we read it, says non-template replies inside the 24-hour customer service window are free, and a free entry point window gives 72 hours of free messages when a user messages through a click-to-WhatsApp ad or Page button.

Some vendors, including WATI, Gupshup and Zoho, have published notices that service messages become billable from October 1, 2026, with 1,000 free per month per number. We could not confirm that on Meta's page, so check it before you budget. Example rates, reported by a third-party blog and not verified on Meta's page, put a US marketing template at $0.0287 and a utility template at $0.0039. Look up your countries on Meta's rate card.

We did not research per-message fees for Instagram, Messenger or website chat, so check each channel's own terms. Our WhatsApp QR vs Business API guide explains how connection choice affects fees and risk.

What does a realistic monthly budget look like?

These are illustrative examples using the numbers above, not quotes. Assume the bot handles all conversations and ignore Meta template fees.

1. Small shop, 500 conversations a month. BossReply Starter at $14 includes 500 bot-driven conversations. A cheap-tier model through your own key adds about $0.90. Roughly $15 a month.

2. Growing coach, 2,000 conversations. BossReply Pro at $29. A cheap-plus model adds about $14.40, a mid model about $72. Roughly $43 to $101 a month.

3. Busy store, 5,000 conversations. Pro at $29 plus a mid model at about $180 comes to roughly $209; with a cheap model it is about $38.

The pattern is clear: model choice moves the total more than the plan does once volume grows, so a model that is good enough is usually a better buy than the best model. Add your own time: a few hours to build the knowledge base and a weekly review.

How can you keep the cost down?

• Start on the cheapest model that passes your own test questions.

• Keep the knowledge base tight. Retrieving fewer, better passages lowers input tokens and improves accuracy. See our knowledge base guide.

• Use fixed rules for simple flows, such as keyword triggers, which need no model call.

• Set a spending cap with your model provider if you use your own key.

• Avoid long minimum terms until you have tested the bot on real conversations.

• Recheck prices each quarter: models and Meta rates change often.

Frequently asked questions

How much does an AI chatbot cost per month?

Platform plans start around $14 for BossReply and go up to hundreds for other tools. Model usage adds roughly $2 to $180 per 1,000 conversations in our worked example, depending on the model.

How much does one AI chatbot conversation cost?

In our example, about $0.0016 to $0.007 with cheap models and $0.036 to $0.18 with mid and top models. The example assumes no caching and no growing history.

Is ChatGPT's API the cheapest option?

Not necessarily. OpenAI, Anthropic and Google each list cheap-tier models at similar prices of about $0.10 per million input tokens. Compare current pages.

Do I pay per message or per conversation?

Model providers charge per token. Platforms vary: per contact, per conversation, per seat, per credit or flat. Compare at your real volume.

Are there hidden WhatsApp costs?

Meta charges for delivered template messages and sets rates by category and country. Vendor notices say service messages may also become billable; verify on Meta's site.

Is bring-your-own-key cheaper?

It can be, because you pay the provider's listed price with no markup from the platform. You will receive a separate bill and should set spending limits.

Does BossReply charge per contact?

No. Plans are not priced by stored contacts. They are $14, $29 and $69 per month.

Can I run a chatbot for free?

Some vendors list free tiers or trials, for example Superchat's free plan for 30 active contacts. BossReply's plans start at $14 per month. Model usage is not free unless a provider offers credits.

Sources and method

Read in October 2026: Anthropic pricing page (platform.claude.com/docs/en/about-claude/pricing) for Claude model prices, cache-read pricing and the batch discount; OpenAI pricing (developers.openai.com/api/docs/pricing), Google Gemini API pricing (ai.google.dev/gemini-api/docs/pricing) and xAI models page (docs.x.ai/docs/models) for per-token prices; Meta WhatsApp pricing documentation (developers.facebook.com/docs/whatsapp/pricing) for the template and service window rules; botpress.com/pricing, stammer.ai/pricing, fastbots.ai/pricing, respond.io/pricing and bossreply.com/pricing for platform plans. ManyChat figures and the example Meta template rates come from a third-party blog (getkanal.com) because manychat.com returned an error. The per-conversation cost model and monthly scenarios are our own arithmetic. We could not verify real-world token counts for your chatbot.

Conclusion

Budget for four layers, not one: the plan, the model, the channel fees and your time. Pick the cheapest model that passes your own tests, compare plans at your real volume and recheck prices every quarter.

To see how the numbers work for your volume, get started with BossReply.