Best voice AI platforms for developers

Last updated: 14 Nov 2026 · Research-based: we compiled public sources; we have not test-called these yet.

Prices change. Every price here is as of 14 Nov 2026 from the vendor's own page. "Not stated" means we couldn't find it on the pages we checked, not that the feature doesn't exist.

Which voice AI platforms are open source?

LiveKit Agents is "open source under the Apache 2.0 license", with Python and Node.js SDKs (LiveKit). Pipecat is "free to use under the BSD-2 license" and is built and maintained by Daily (Pipecat). Both run on your own servers or on the company's cloud.

Everything else here is a hosted service. Bland is the only one that lists on-prem or VPC deployment, on its Enterprise plan (Bland).

How do the pricing models compare?

• Platform fee plus pass-through costs: Vapi and Retell. You pay their fee, then speech-to-text, LLM, voice and the phone line.
• One bundled rate: Bland's per-minute price covers LLM, speech-to-text and voice, with "No token charges"; telephony is separate.
• Plans with included minutes: ElevenLabs Agents and LiveKit Cloud.
• Hosting per agent minute: Pipecat Cloud, with models billed by each provider.
• Tokens: OpenAI Realtime bills audio and text tokens, so the cost of a minute depends on how much is said.
• Orchestration per minute: Twilio ConversationRelay, plus Twilio voice minutes.

Our AI voice agent pricing guide works through all-in costs per minute.

The platforms compared

LiveKit Agents

• Open source: yes, Apache 2.0.
• Pricing (as of 14 Nov 2026): the framework is free. LiveKit Cloud's Build plan is $0 with 1,000 agent session minutes and 5 concurrent sessions; Ship starts at $50/month with 5,000 minutes, then $0.01/min; Scale starts at $500/month with 50,000 minutes and up to 600 concurrent sessions (LiveKit).
• Telephony: SIP for placing and receiving calls. On Cloud, every plan includes 1 free US local number. On Ship, extra numbers cost $1/month and third-party SIP minutes cost $0.004/min after the included 5,000.
• Models: plugins for "most major AI providers" across LLM, speech-to-text, text-to-speech and realtime APIs; LiveKit Inference offers models without separate API keys.
• SDKs: Python and Node.js.
• Self-host: yes, with self-hosted LiveKit servers and Kubernetes support.
• Watch out: you write and run the agent code, and the pricing page lists a "Signed BAA" on Enterprise only.

Pipecat

• Open source: yes, BSD-2, maintained by Daily.
• Pricing (as of 14 Nov 2026): the framework is free and you pay for the AI services you choose (Pipecat). Pipecat Cloud hosting costs $0.01/min per active agent on its smallest size, up to $0.03/min on the largest (Daily).
• Telephony: Daily phone dial-in and dial-out cost $0.018/min with SIP included, Daily SIP alone is $0.003-$0.02/min, and call transfers cost $0.20 each. Twilio or your own WebSocket provider bills you directly.
• Models: the docs say it "works with any AI provider", and the vendor says it can orchestrate "150+ AI services".
• SDKs: a Python framework, with client SDKs for JavaScript, React, React Native, iOS, Android and C++.
• Self-host: yes, or use Pipecat Cloud.
• Watch out: you assemble more yourself, and the Pipecat Cloud pricing page doesn't mention HIPAA or SOC 2.

Vapi

• Open source: no, it's a hosted platform.
• Pricing (as of 14 Nov 2026): $0.05/min hosting, with providers passed through "with no markup" (Deepgram $0.0095-$0.0099/min, OpenAI $0.0077-$0.0452/min, ElevenLabs $0.0146-$0.0238/min). Core is $29/month; Pro is 10% of the hosting fee with a $999/month minimum. $5 in free credits; extra concurrency $10 per line per month (Vapi).
• Telephony: Vapi Telephony and SIP, WebSockets and Daily WebRTC are free. Twilio ($0.008/min inbound, $0.014/min outbound), Vonage ($0.00814/min) and Telnyx ($0.0055/min) are charged by those providers.
• Models: a "Full model catalog" at cost. SDKs aren't listed on the pricing page.
• Self-host: not offered on the pricing page.
• Watch out: HIPAA is a $2,000/month add-on, and the headline rate leaves out model and phone costs. See Best Vapi alternatives and Retell AI vs Vapi: pricing, compliance and fit compared.

Retell

• Open source: no.
• Pricing (as of 14 Nov 2026): $0.07-$0.31/min, built from Retell Voice Infra at $0.055/min, voice at $0.015/min (or $0.040-$0.10/min for ElevenLabs voices), an LLM from $0.0016/min (GPT 5 nano) and telephony. $10 in free credits; 20 concurrent calls free, then $8 per concurrent call per month; billed to the second (Retell).
• Telephony: Twilio or Telnyx at $0.015/min in listed countries; no charge for SIP trunking or custom telephony; numbers $2/month.
• Models: OpenAI, Anthropic and Google LLMs, speech-to-speech options (GPT Realtime, Grok Voice) and a "Custom LLM" option. Voices from Retell, Minimax, Fish, Cartesia, OpenAI, Inworld and ElevenLabs.
• SDKs and self-host: SDKs aren't mentioned on the pricing page; no self-hosting, though Enterprise offers a dedicated server add-on.
• Watch out: add-ons stack (knowledge base +$0.005/min, PII removal +$0.01/min), and prompts averaging over 4,000 tokens increase billed time.

ElevenLabs Agents

• Open source: no.
• Pricing (as of 14 Nov 2026): Free (15 minutes), Starter $6/month (75 minutes), Creator $22 (275), Pro $99 (1,238), Scale $299 (3,738), Business $990 (12,375). The page lists $0.080 per extra minute and $0.160 per burst minute; LLM cost is "Based on usage and varies by model" (ElevenLabs).
• Telephony: "At cost".
• Models: you choose the LLM and pay for usage. SDKs and bring-your-own-LLM options aren't covered on the pricing page, which links to an Agents API.
• Self-host: not offered.
• Watch out: LLM and telephony sit outside the plan price, and BAAs are Enterprise-only.

OpenAI Realtime API

• Open source: no.
• Pricing (as of 14 Nov 2026): gpt-realtime costs $32 per 1M audio input tokens ($0.40 cached), $64 per 1M audio output tokens, $4 per 1M text input tokens and $16 per 1M text output tokens (OpenAI).
• Telephony: the model responds "over WebRTC, WebSocket, or SIP connections".
• Models: 1 speech-to-speech model listens and speaks, so there's no separate voice engine to swap. Function calling is supported.
• SDKs and self-host: not covered on the model page; no self-hosting.
• Watch out: the original snapshot (gpt-realtime-2025-08-28) is marked deprecated, and newer realtime models exist (Retell's price list includes GPT Realtime 2 and 2.1), so check OpenAI's current model and price before you build.

Twilio ConversationRelay

• Open source: no.
• Pricing (as of 14 Nov 2026): $0.07/min pay-as-you-go; "Voice costs are calculated separately" (Twilio).
• Telephony: Twilio Voice. You add <ConversationRelay> to TwiML with a wss:// URL for your WebSocket server (Twilio docs).
• Models: Twilio runs speech-to-text (Deepgram, Google) and text-to-speech (Google, Amazon Polly, ElevenLabs), and your server decides what to say, so the LLM is your choice.
• SDKs: Twilio's helper libraries; the docs example uses Node.js.
• Self-host: not offered, though your WebSocket app runs wherever you like.
• Watch out: you build and host the conversation logic, and the pages we checked don't mention HIPAA.

Bland

• Open source: no.
• Pricing (as of 14 Nov 2026): Start has no platform fee, $0.14/min talk time, $0.05 per transfer minute, 10 concurrent calls and 100 calls a day. Build is $299/month with $0.12/min, $0.04 per transfer minute, 50 concurrent calls and 2,000 calls a day. Enterprise is custom (Bland).
• Telephony: bring your own Twilio account or SIP trunks and pay your carrier, or use Bland's built-in Twilio at pass-through cost. Bring-your-own-telephony customers don't pay transfer fees.
• Models: bundled; the page doesn't mention bringing your own LLM or SDKs.
• Self-host: "On-Prem/VPC available" on Enterprise.
• Watch out: daily call caps on Start and Build, and the BAA is Enterprise-only. See Best Bland AI alternatives.

Making outbound calls with any of these? Read Are AI phone calls legal? US, UK, EU and India first.

How we picked

• Developer access: an API or SDK, phone connectivity and a choice of models.
• Public pricing, re-checked on 14 Nov 2026.
• A mix of open-source frameworks and managed APIs, since the right pick depends on how much you want to run yourself.
• No speed or call-quality claims until we've made our own calls. Full method: How Kall tests and ranks AI voice agents.

Sources

1. Pricing, Vapi, accessed 14 Nov 2026
2. Pricing, Retell AI, accessed 14 Nov 2026
3. Conversational AI solutions pricing, Twilio, accessed 14 Nov 2026
4. Agents introduction, LiveKit docs, accessed 14 Nov 2026
5. Pricing, LiveKit, accessed 14 Nov 2026
6. Introduction, Pipecat docs, accessed 14 Nov 2026
7. Pipecat Cloud pricing, Daily, accessed 14 Nov 2026
8. ElevenAgents pricing, ElevenLabs, accessed 14 Nov 2026
9. GPT-Realtime model, OpenAI, accessed 14 Nov 2026
10. ConversationRelay, Twilio docs, accessed 14 Nov 2026
11. Pricing, Bland AI, accessed 14 Nov 2026

FAQ

Related

• AI voice agent pricing per minute: the real all-in costWhat AI voice agents and receptionists really cost per minute once platform, LLM, voice, transcription and phone line are added. 15 platforms.
• Best Vapi alternatives7 Vapi alternatives with pricing as of 14 Nov 2026: Retell AI, Bland, ElevenLabs, Deepgram, PolyAI, LiveKit and Pipecat, with a watch-out for each.
• Retell AI vs Vapi: pricing, compliance and fit comparedRetell AI vs Vapi on per-minute cost, free credits, HIPAA and SOC 2, integrations, no-code fit and languages, from official pages as of 14 Nov 2026.
• Best Bland AI alternatives7 Bland AI alternatives with pricing as of 14 Nov 2026, why you might switch from Bland's $0.14/min rate, and what to watch out for with each.

Ranks are never for sale. This page has no affiliate links and no paid placements. Prices change often: check the vendor's pricing page before you buy.