chatbots
By Saurav | Founder of saavos | Building in public toward $10k MRR
[!TLDR] AI chatbot prices look similar on the surface — most start at $19–49/month — but the underlying model, message volume, and billing model differ enough that the same dollar buys wildly different products. This is the honest pricing comparison across the six platforms that cover most of the buyer journey in 2026: Chatbase, Intercom Fin, Tidio, Crisp, Botpress, and saavos. Skip the feature checklists. Read this if you want to know what you're actually paying for.
Four things, in different proportions: the underlying language model, inference volume, infrastructure reliability, and vendor margin. The model is the most important and least advertised. GPT-3.5 versus Claude Sonnet 4.6 isn't a minor preference — it's a 3–5× cost difference per token, which is why two "$19/month" plans can have wildly different message caps. Volume requirements depend on traffic, not vendor brand.
When you pay $19/month for a chatbot, you're paying for four things in different proportions: the underlying language model, the inference volume, the infrastructure to run it reliably, and the company's margin. Vendors don't break this down on their pricing pages, but the differences are huge.
A platform running GPT-3.5 can sell you 5,000 messages for $19 and still profit. A platform running Anthropic Claude Sonnet 4.6 — roughly 3–5× more expensive per token — has to either cap your volume tighter or charge more. Neither is "wrong"; they're answering different buyer questions. A solo founder with 200 FAQ-style questions a month is overpaying for Sonnet. A founder taking high-stakes sales conversations through their bot is underspending on GPT-3.5.
The pricing comparison below is structured to surface those trade-offs. We list the six platforms most pre-revenue and early-revenue founders evaluate, what each charges, what's actually included, and the hidden costs that don't show on the marketing page. If you want a working bot live in 5 minutes after reading this, you can start free on saavos — but read the rest first.
Entry-tier paid pricing in 2026: Chatbase $40/mo (monthly) or $32/mo (annual), saavos $19/mo, Tidio $39/mo (Lyro AI tier), Botpress pay-as-you-go ($0.075/AI message over 1k free), Crisp ~$103/mo (Chatbot Plus, EUR-priced), Intercom Fin $99/mo base plus $0.99 per resolved conversation. Free tiers exist on most but cap out at 50–100 conversations — enough to evaluate, not to run.
| Platform | Entry price (paid) | Free tier | Messages at entry tier | Underlying model | Billing model |
|---|---|---|---|---|---|
| Chatbase | $40/mo (monthly) / $32/mo (annual) | Yes (very limited) | ~2,000 | GPT-3.5; GPT-4 on higher | Flat subscription, message overages |
| Intercom Fin | $99/mo + $0.99 per resolution | No | Pay-per-resolved | OpenAI | Per-resolution on top of platform contract |
| Tidio | $39/mo (Lyro AI) | 50 Lyro convos (one-time) | 100 | Lyro proprietary | Live chat + bot bundle, separate AI tier |
| Crisp | ~€95/mo (Chatbot Plus) | Limited bot only | Varies | Flow-builder + GPT add-on | Live chat platform + bot upcharge |
| Botpress | Pay-as-you-go (~$0.075/AI msg over 1k free) | 1,000 AI msgs/mo free | Variable | Dev-configurable (Claude, GPT) | Usage-based metered |
| saavos | $19/mo | 50 conversations/mo, forever | ~1,000 | Claude Sonnet 4.6 (all paid tiers) | Flat subscription, no per-message charges |
A few footnotes the table can't carry:
A useful third-party reference is Intercom's own Fin pricing page, which confirms the per-resolution model and the platform-required structure. Read it next to ours and the picture sharpens.
ALT: Side-by-side bar chart of monthly cost at 1,000 conversations across six chatbot platforms — saavos $19, Chatbase $40 (monthly), Botsonic $19+overages, Tidio $79, Intercom Fin $99+platform, Crisp ~$103.
It depends on the vendor. At $40/month (monthly) or $32/month (annual), Chatbase gives you 2 bots and ~2,000 messages on GPT-3.5. saavos gives you 2 bots and ~1,000 messages on Claude Sonnet 4.6 at $19/month — half the volume but a meaningfully stronger model and $13–$21/month less. Intercom Fin starts at $99/month base plus $0.99 per resolved conversation, which is unpredictable. Tidio at $39/month gives you 100 Lyro AI conversations — not a typo. Read each tier breakdown below.
Two bots, 4 million training characters, ~2,000 messages/month. GPT-3.5 by default. Pricing scales linearly into Standard ($49) and Pro ($99) tiers, which include GPT-4 and more bots. Where Chatbase wins on price: their template library lets you skip the "what should my bot do?" decision. Where it loses: the GPT-3.5 default on the entry tier is meaningfully weaker than frontier models on multi-turn or ambiguous queries.
Resolution-based billing is genuinely unpredictable. A "resolution" is a conversation Fin handles without escalation. If your bot handles 200 conversations in a slow month, that's $99 + $198 = $297. If it handles 800, that's $99 + $792 = $891. There's no upper bound. For a team already on Intercom this is fine — it's variable cost that scales with revenue. For a solo founder evaluating chatbots cold, the unpredictability is a real problem.
Two products fused into one — the Tidio live-chat suite (free or $29/mo for Communicator) and the Lyro AI bot ($39/mo for 100 conversations). To get a bot answering 1,000 questions/month you're at the $79+/mo Lyro tier. Tidio shines on ecommerce: Shopify product lookups, abandoned-cart triggers, order status. If you don't sell physical goods, you're paying for features you won't use.
Live chat first, chatbot bolted on. The flow-builder is genuinely good (visual, branching, no-code) but the GPT-add-on for AI responses is comparatively basic — it doesn't have purpose-built retrieval grounding the way Chatbase, saavos, or Wonderchat do. If you mainly need a multi-channel inbox (chat + WhatsApp + Messenger + email) and want a bot as a side feature, Crisp is well-positioned for that price.
Free tier: 1,000 AI messages/month for as long as you keep building. Past that, $0.075/AI message ≈ $7.50 per additional 100 messages. The platform itself is free. Where Botpress wins: maximum flexibility, self-host option, channel sprawl (WhatsApp, Telegram, Slack, Teams, etc.). Where it loses: you're a developer building a bot, not a buyer of a finished bot. If your time-to-deploy budget is 5 minutes, this isn't your tool.
Two bots, 10 sources per bot, ~1,000 conversations/month. Powered by Claude Sonnet 4.6 (the default on all paid tiers; Free uses Haiku). Flat subscription — when you hit your cap, the bot pauses politely until next billing period. No per-resolution surcharge, no overage on the Starter or Pro tiers. Dodo Payments handles billing as Merchant of Record, which means tax and invoicing are sorted for international customers automatically. Where we win: speed-to-live (5 minutes from URL paste), Anthropic-grade model quality, predictable bill. Where we lose: no live-chat inbox, fewer integrations than Tidio or Intercom, and no on-premise option for buyers who need self-hosted.
Three categories most buyers miss: per-resolution billing creep (Intercom Fin's $0.99/conversation compounds fast — 800 resolved conversations in a good month is $891 total), tax handling (non-US buyers need a Merchant of Record or face manual VAT compliance), and onboarding time (a platform that takes 2 hours to configure costs $200+ of your time before it answers one question). All three are invisible on pricing pages.
If you're evaluating Intercom Fin (or any per-resolution model), do the math against your actual conversation volume — not the marketing average. Pull last month's customer support volume, multiply by your expected deflection rate (assume 60% to be conservative for a new bot), then multiply that by the per-resolution fee. The honest number is usually 2–3× the marketing-page sticker.
If you're outside the US, your chatbot vendor's invoicing setup matters. Some vendors are Merchant of Record (MoR) — they collect tax, remit it to your government, and issue you a compliant invoice. Others are not, which means you're responsible for figuring out reverse-charge VAT, GST registration, or the equivalent. Stripe-based vendors are usually not MoR. Paddle, Dodo, and Lemon Squeezy are. This isn't visible in the price comparison but it's hours of accountant time annually.
If a chatbot takes 30 minutes to set up versus 5, you've spent $50–100 of your hourly time before the bot answers a single question. Multiply that by how many bots you're evaluating and the implicit cost of "free trials" is real. The $19/mo platform that's live in 5 minutes is actually cheaper than the $19/mo platform that takes 2 hours to configure — even though the price is identical.
What happens when your bot has a great Reddit-front-page day and gets 10× normal traffic? On flat-subscription platforms (saavos Starter, most Chatbase tiers), the bot pauses at the cap. On per-message platforms (Botpress, some overage modes), you wake up to a bill. Neither is wrong, but you should know which mode you're in. We deliberately designed saavos's Starter and Pro tiers to never produce a surprise bill — see our pricing page for the exact caps.
Estimate monthly conversation volume, then buy the tier that covers it with 30% headroom. Pre-launch sites under 100 visitors a day need 30–80 conversations a month — free tier on any platform. Early-revenue SaaS at 100–500 visitors a day needs 200–800 conversations — Starter tier. Established SMBs at 500–2,000 visitors a day need 1,000–3,000 — Pro tier. Most teams overestimate by 3–10x.
A simple heuristic: estimate your monthly conversation volume, then pick the tier that covers it with 30% headroom. Don't pay for 15,000 messages when you'll use 800. Don't buy 100 messages when you'll need 2,000 and hit the cap weekly.
How to estimate volume:
These numbers assume roughly 5–8% of visitors open the chatbot. If you have an active community or a high-intent traffic source, multiply by 1.5–2×. If your traffic is mostly bounce-rate, divide by 2.
Three ways: free tiers capped so low you upgrade in week one anyway (50 messages isn't a real test), premium tiers that bundle enterprise features (SSO, SOC 2, audit logs) you won't need for years, and "almost-included" features that are actually paid add-ons — live agent handoff, white-label branding, and API access are all frequent upcharges on top of the listed price.
Several platforms advertise "free forever" but the actual free tier is too thin to evaluate the product seriously — 50 messages one-time, no widget customization, the vendor's branding plastered across the launcher. If you have to upgrade in week one just to test the bot, the free tier was marketing, not a product. Test free tiers against the question: "can I leave this on my live site for a month without embarrassment?" If no, it's a trial dressed as a tier.
On the other end: $199–499/mo "Business" or "Scale" tiers that include features (SSO, custom DPA, audit logs) that don't matter until you have a security-conscious enterprise buyer. Solo founders shouldn't pay for SOC 2 Type II until a customer asks. Match tier to actual need, not aspirational need.
Watch for features listed on the pricing page that are actually paid add-ons. Common ones: live-agent handoff (often a $20–50/mo add-on on top of the bot subscription), white-label branding (frequently Pro+), API access (usually Business+). The sticker price is rarely the all-in price for the configuration most buyers actually want.
Three questions: Do you already pay for Intercom, Zendesk, or Help Scout? If yes, evaluate their AI add-on before buying something separate. Are you selling physical goods? If yes, Tidio's ecom features (Shopify lookups, abandoned-cart triggers) are hard to replace. If neither, and you want a bot live today with predictable billing — saavos at $19/month or Chatbase at $32–$40/month. The $13–$21 gap gets you a more established template library and WhatsApp/Instagram channels; if you don't need those, it doesn't.
Three questions, in order:
That's it. Three questions narrow the six platforms down to one or two real candidates. The rest is feature trial.
Because per-resolution billing is unpredictable for pre-revenue founders. A good Reddit day can send your Intercom Fin bill from $100 to $890 in a single month. saavos's Starter and Pro tiers both pause at the cap — no overage. Worst case you know your maximum bill before signing up. We chose flat subscription because our ICP needs predictable burn, not variable cost tied to traffic luck.
We chose flat subscription instead of per-resolution because the alternative is bad for our ICP. Pre-revenue founders need predictable burn. A bot that costs $19/mo or $890/mo depending on Reddit luck isn't a tool — it's a liability. Our Starter pauses at the cap. Pro pauses at the cap. Only Business has overage, and even that's capped at $200/period (overage billing is in beta; current overages are absorbed by saavos until metered billing ships, target 2026-06). Worst case: you know your maximum bill before you sign up.
We chose Anthropic Claude Sonnet 4.6 as the default model because it's measurably better at the multi-turn, ambiguous queries real visitors send — and we'd rather charge $19 for a bot that answers correctly than $9 for one that hallucinates. The Free tier is Haiku, which is still excellent at FAQ deflection. The model choice is per-bot configurable on Pro and Business, so power users override.
We chose Dodo as Merchant of Record because we hate the alternative. Manually figuring out VAT for our first international customers was an avoidable pain. Dodo charges a slightly higher take rate than Stripe, and we pay the difference so you don't have to think about it.
If those trade-offs sound right, start free on saavos — 50 conversations forever, no card. If they don't, the comparison above is honest about which platform fits you better. Pick the right tool, not ours.
Ten things: actual monthly conversation volume, tier with 30% headroom, underlying model on the entry tier, all-in cost including add-ons, billing model (flat or per-resolution), Merchant of Record status if you're outside the US, free-tier test with one real-world question, at least one critical review from a non-vendor source, embed snippet working on your CMS, and a cancellation reminder set 24 hours before trial ends.
If you want a head-to-head on specific platforms, our comparison pages cover the head-to-heads in detail. Or see our pricing for the four tiers + what each one buys you.
Before you commit to a platform, it's worth running through the 12 questions every buyer should ask AI chatbot vendors — they surface the billing edge cases and contract terms that don't show on pricing pages. If you've already narrowed to Chatbase or Intercom Fin specifically, our Chatbase alternatives guide and Intercom Fin alternatives guide compare each platform against its real-world substitutes.
ALT: Decision-tree diagram — "Do you already use Intercom/Zendesk? -> Use their AI. Otherwise: physical goods? -> Tidio. Digital and want 5-min setup? -> saavos/Chatbase. Need maximum flexibility? -> Botpress."
Get the next post in your inbox
Honest writing on building, embedding, and shipping AI chatbots. No spam. Unsubscribe anytime.
The underlying model is the biggest cost driver. Anthropic Claude Sonnet 4.6 costs roughly 3–5× more per token than GPT-3.5. A platform shipping GPT-3.5 on the $19 tier can include 5,000 messages and still profit; a platform shipping Sonnet 4.6 has to cap closer to 1,000. Pick by what your visitors ask — Sonnet wins on multi-turn or ambiguous queries; GPT-3.5 is fine for high-volume FAQ deflection.
Fin is $99/seat/month for the Intercom platform plus $0.99 per resolved conversation. The headline $0.99 isn't the headline number — the all-in cost for a 3-person team handling 500 conversations a month is closer to $600–800 (3 seats + 500 resolutions). For pre-revenue founders, the unpredictability is a bigger problem than the absolute number: a viral Reddit day can spike your Intercom bill 5–10×. Flat-subscription platforms pause at the cap instead.
Four costs vendors don't list. (1) Per-resolution billing creep — multiply your actual deflection volume by the per-resolution fee; the all-in is usually 2–3× the marketing sticker. (2) Tax handling — non-Merchant-of-Record vendors leave you responsible for VAT, GST, and reverse-charge compliance. (3) Onboarding time — a platform that takes 30 min to set up vs 5 min costs you $50–100 of your hourly time before the bot answers anything. (4) Overage on burst traffic — some platforms charge for surprise spikes; flat-subscription platforms pause instead.
Match tier to actual monthly conversation volume with 30% headroom. Under 100 conversations/month: free tier. 100–700/month: $19 Starter. 700–2,500/month: $49 Pro. 2,500–10,000/month: $199 Business (metered overage above 15k). All paid tiers run Sonnet 4.6; Free runs Haiku 4.5. Don't pay for 15,000 messages when you'll use 800; don't buy 100 messages when you'll hit the cap weekly. Pull your actual monthly support volume, multiply by your expected deflection rate (~60%), and that's the message count you need.
For pre-revenue and early-revenue founders: flat subscription wins on predictability. The worst-case bill is the plan price (plus capped overage on Business-tier plans). For established teams with steady support volume and existing Intercom or Zendesk infrastructure: per-resolution can be cheaper at scale. The pivot point is roughly 2,000+ conversations/month sustained for 12+ months with a dedicated support team. Below that, flat-subscription removes a real source of founder anxiety.
Builds tools for solopreneurs and small SaaS teams who don't have an afternoon to spare.
Paste your URL. Train your bot. Drop one script tag. No credit card.