The twelve platforms that turn an LLM into a phone call — what they are, what they cost, where they break, who they're for. Tap any one for the full page.
A voice agent is assembled from four layers. Most "which platform is best?" arguments are really arguments between layers — knowing which one your problem lives in narrows twelve options to two or three.
Deepgram / ElevenLabs Scribe (hear) · OpenAI / Claude / Gemini (think) · ElevenLabs / Cartesia (speak) — or a native speech-to-speech model (OpenAI GPT-Realtime, Gemini Live). Note: Claude has no native speech-to-speech, so a "Claude voice agent" is always the stitched STT→Claude→TTS path.
You design the flow, pick models, connect telephony. Days-to-weeks to ship, transparent per-minute pricing. The DIY / agency layer — and where an Articulate-built agent lives.
The WebRTC/SIP plumbing and voice engine the platforms themselves run on. Use directly only if you're building your own platform or need full data-sovereign control.
Done-for-you, vertical, contract-priced ($150K–$1M+/yr). You buy an outcome, not a builder.
Tap a card for the full page — what it is, key features, where it breaks, real use cases, pricing, and Articulate's take. build · models/infra · managed
Pricing transparency: clear mixed opaque. Rows are tappable.
| Platform | Best for | Latency | Pricing | Ship | Edge |
|---|
Vendor-stated or third-party-reported, June 2026. Verify against a live pilot before committing.
The honest match. Tap a tool chip to open its full page.