Conversational AI Platforms Compared (2026)

Conversational AI platforms are the tools you use to build an agent that talks to people, by chat or by phone, and then plug it into your CRM, calendar and help desk. If you want a phone agent and have a developer, start with Vapi or Retell. If you want a support bot on top of an existing help desk, look at Intercom Fin. If your company already lives in Google Cloud, AWS or Microsoft 365, the native option is usually cheaper than it looks. I checked every price below on the vendor's own pricing page in October 2026, and I'll tell you which one runs the chat assistant on this website.

What kinds of conversational AI platforms are there?

Search for this term and you get products that do very different jobs, so I sort them into four groups before comparing prices.

Voice-first infrastructure (Vapi, Retell, ElevenLabs Agents, Bland) gives you a phone or web agent and bills by the minute. You bring the prompt, the tools and usually the phone number.

Agent builders (Voiceflow, Botpress) give you a visual canvas for designing chat and voice agents, with your choice of language model underneath.

Support agents (Intercom Fin) answer customer questions from your help center and hand off to a human. You configure them more than you build them.

Cloud suites (Google Conversational Agents, Amazon Lex, Microsoft Copilot Studio) are the big-cloud versions. They're priced per request or per credit and make the most sense when your data and identity already sit in that cloud.

Gartner's Peer Insights category covers the enterprise contact center end of this market (Cognigy, Kore.ai, Yellow.ai, PolyAI). Those are sold through sales teams with custom quotes, so I left them out. Everything below has a price you can see before talking to anyone, or says plainly that it doesn't.

How do conversational AI platforms compare on price?

All figures are from each vendor's pricing page, checked October 2026.

Per-minute prices on voice platforms are the hardest to compare, because each one bundles different pieces. Bland's $0.14 includes the model. Vapi's $0.05 doesn't. Read the right-hand column before you read the middle one.

Voice-first platforms

Vapi

Vapi charges a $0.05 per minute hosting fee and passes through the model, transcriber and voice costs from whichever providers you pick, which it says it does with no markup. Its own estimate for 1,000 minutes with Deepgram, an OpenAI model and an ElevenLabs voice comes to $82 to $129 a month, so plan on roughly $0.08 to $0.13 a minute all in. The free tier gives you $5 of credit and 4 concurrent calls. Paid support packages start at $29 a month (10 concurrent calls) and the Pro package has a $999 monthly minimum. HIPAA is a $2,000 a month add-on, and extra concurrent lines are $10 each per month.

Vapi is the most configurable option here and it's built for developers. You'll be writing prompts, tool definitions and webhooks, and owning the integration work. More on that below, because it's what we run.

Retell AI

Retell prices voice agents at $0.07 to $0.31 a minute depending on the model and voice you choose, built from a $0.055 per minute voice engine plus the LLM and text-to-speech you select. Chat agents start at $0.002 per message. New accounts get $10 of free credit and 20 concurrent calls, and each concurrent call past that is $8 a month. Add-ons are itemized: knowledge base is $0.005 a minute, PII removal $0.01 a minute. Retell says it's SOC 2 certified and HIPAA-ready.

Retell is the closest competitor to Vapi and the pricing page is the most transparent in this list. If you want a developer platform with less assembly than Vapi, start here. Our Bland vs Vapi vs Retell breakdown goes deeper on the three.

ElevenLabs Agents

ElevenLabs sells its agent product on subscription tiers with call minutes included: 15 minutes free, 275 minutes on the $22 Creator plan, 1,238 on the $99 Pro plan, up to 12,375 on the $990 Business plan. Extra minutes are $0.08 on every tier and text messages are $0.003 each. LLM usage is billed separately by model, and telephony from your own provider is at cost. HIPAA BAAs are Enterprise only.

Pick it for the voice quality. ElevenLabs built its business on speech synthesis, and other platforms resell its voices (Retell lists them as a paid upgrade on its own pricing page).

Bland AI

Bland is the simplest to price: $0.14 a minute with no platform fee, or $299 a month plus $0.12 a minute on the Build plan. That one rate covers the LLM, speech-to-text and text-to-speech. The catch is the caps. The free Start plan allows 10 concurrent calls and 100 calls a day, and Build raises that to 50 concurrent and 2,000 a day. If you want a phone agent without stitching providers together, Bland is the shortest path.

Agent builders for chat and voice

Voiceflow

Voiceflow no longer publishes plan prices. Agencies get a free trial and usage-based billing, and businesses are asked to book a demo. It supports chat and voice, lets you choose between the major model providers or bring your own, and has proper environments for moving an agent from development to staging to production. It's a good fit for a product or CX team that wants to design conversations visually and needs non-developers to edit them.

Botpress

Botpress has a free plan with 100 conversations, 3 seats and 3 agents. The first paid plan is $150 a month for 250 conversations with $25 of AI usage included, and the Team plan is $750 a month for 1,500 conversations, $150 of AI usage and unlimited seats. Botpress says every conversation includes $0.10 of AI usage it covers. If your bot runs long conversations on an expensive model, the AI line is the one to watch.

Support agents

Intercom Fin

Fin costs $0.99 per outcome, and Intercom says you pay nothing if Fin can't answer, fails an action, or the customer asks for a human. It works with any help desk, including Zendesk, Salesforce and Freshdesk, so you don't have to move off your current one. Fin Voice is sold through sales at $1.99 per voice outcome. There's a 14-day free trial.

Outcome pricing suits support, because you only pay when a question gets answered. For sales or lead capture it fits badly, since a "resolution" is hard to define there. If your main job is deflecting support tickets, Fin is the first thing I'd trial.

Cloud suites

Google Conversational Agents (Dialogflow CX)

Google now sells Dialogflow CX as Conversational Agents. Chat costs $0.007 per request with scripted flows and $0.012 with generative playbooks. Voice costs $0.001 a second with flows and $0.002 with playbooks, which works out to $0.06 and $0.12 a minute. New users get $600 of flow credit and $1,000 of playbook credit that expire after 12 months.

Amazon Lex

Amazon Lex charges $0.00075 per text request and $0.004 per speech request. That's the cheapest per interaction on this list, but you're paying for the language layer only. The phone line usually comes from Amazon Connect and the business logic from Lambda, both billed separately. Lex makes sense when your team already builds on AWS.

Microsoft Copilot Studio

Copilot Studio is sold as a tenant-wide license with capacity packs of 25,000 Copilot Credits for $200 a month, or on a pay-as-you-go meter. Users licensed for Microsoft 365 Copilot can build and use internal agents at no extra cost. It's the obvious choice for internal agents in Teams. For a customer-facing website or phone agent, the platforms above are more direct.

What do we run on our own site?

The chat assistant on whitespacesolutions.ai is built on Vapi, in chat mode. The widget runs in chat mode, so it produces no calls at all. That changes how leads arrive. A Vapi chat reports through session.updated events instead of the end-of-call-report a phone call sends, and a chat session reports many times as it grows. Our webhook keys the lead on the session id, updates that one row as the chat grows, and sends the n8n notification once, on the first report that carries contact details.

Running it taught me three things that apply to any platform you pick.

Manage the agent as code. The prompt, the knowledge base and the server settings live in our repo, and a script applies them to Vapi and a second script fails if the live agent drifts from the repo. We do this because saving the assistant in Vapi's dashboard re-serialized the whole config and silently overwrote settings we'd applied through the API. If your platform has a dashboard and an API, pick one as the source of truth.

Use structured outputs for lead data. Vapi extracts the lead profile, a summary, whether an appointment was booked, and a qualification score as structured fields, and our webhook reads those directly. Parsing a transcript yourself is slower and breaks every time the conversation changes.

Store every conversation before you do anything else with it. Every chat lands in a Postgres table with its transcript and score, and only conversations that produce an email or phone number become leads. Vapi doesn't retry failed webhook deliveries unless you configure it, so we set three retries. Before that, one server error meant a lost conversation. Also check the list of events your webhook subscribes to, because leaving one out produces no error and no deliveries for that event.

None of that is unique to Vapi. It's the work between a demo agent and one you trust with leads, and it's why the per-minute price is only a small part of what an agent costs. For the real estate version, including what it should ask sellers, see buying or building a real estate chatbot.

How should you choose a platform?

Start with the channel. If the agent answers or makes phone calls, you're choosing between Vapi, Retell, ElevenLabs and Bland. If it's website chat, every platform on this list can do it, and the question is what it connects to.

Then look at who will maintain it. Vapi and Retell assume a developer. Bland, Fin and Copilot Studio can be run by an ops person. Voiceflow and Botpress sit in between.

Then price your actual volume. Take last month's call minutes or chat conversations and multiply them out on two or three platforms. At 1,000 minutes a month the differences are tens of dollars, and at 50,000 they decide the budget.

Finally, check where the data goes. If you handle health information, the HIPAA cost ranges from an add-on (Vapi at $2,000 a month) to an Enterprise contract (ElevenLabs). If you need transcripts in your own database, confirm the platform sends them by webhook and test it before you sign anything.

For receptionist-style tools that come ready to use rather than built, see our AI receptionist comparison.

What rules apply to agents that make phone calls?

If your agent calls or texts people, you're responsible for following the law whichever platform you use. A few rules apply to every vendor above. This is not legal advice.

The FCC ruled in February 2024 (FCC 24-17) that AI-generated voices are "artificial voices" under the TCPA. Calls made with them need the recipient's prior express consent, and written consent for telemarketing. Ringless voicemail counts as a call. The agent's script should say it's an AI at the start of the call.

Marketing texts need prior express written consent. A2P 10DLC registration has been required for business texting from local numbers since February 1, 2025, and registering is not consent. We cover calling rules in AI vs human cold callers and texting rules in AI SMS compliance.

On our own site, SMS consent is stored in its own database columns, and all of them default to false. A missing opt-in reads as a refusal, which is the only safe default when a carrier or a lawyer asks who agreed to what.

If you'd rather have the agent built and wired into your CRM than assemble it yourself, that's our AI chatbot development work, and AI voice agents for the phone side.