LiveKit Agents $0.01 a minute. Pipecat Cloud $0.01. Voximplant $0.004. All three have a free tier and self-serve signup, all read 2026-08-07.
The spread is six tenths of a cent, which on 10,000 minutes a month is $60. That is not a decision. The decision is what each one leaves you holding.
What the rate covers, and what it does not
| LiveKit Agents | Pipecat Cloud | Voximplant | |
|---|---|---|---|
| Published rate | $0.01/min | $0.01/min | $0.004/min |
| Recognition | Excluded | Excluded | Excluded |
| The model | Excluded | Excluded | Excluded |
| Synthesis | Excluded | Excluded | Excluded |
| Telephony | Excluded | Excluded | Included in rate |
| Free tier | Yes | Yes | Yes |
| Self-serve | Yes | Yes | Yes |
All three are transport and orchestration. None supplies the parts that make an agent talk. Add published list rates for recognition, a low-latency model and synthesis and the floor lands near $0.0465 a minute whichever of the three you pick.
The real difference is what you maintain
LiveKit Agents is a framework, not a product. There is no flow editor and no non-technical path. It hands you turn detection, barge-in and reconnection as your problems, on top of the WebRTC infrastructure a large share of this category already runs on.
Pipecat Cloud removes hosting toil, not integration work. It is Daily's hosted runtime for the open-source Pipecat framework, which is the orchestration layer several other platforms are quietly built on. You still assemble and pay for every component.
Voximplant is lower-level again, with its own agent layer for developers who want call control. More code, more control, and telephony already inside the rate rather than billed by a carrier alongside it.
When any of these beats a platform
A managed platform charges $0.05 to $0.11 a minute for orchestration plus components. These three charge a cent or less for transport. The gap is not margin, it is turn-taking and interruption handling that took years to tune.
- You have engineers who will own turn detection and reconnection, and want to.
- You need call-control behaviour a platform's flow editor cannot express.
- Your volume is high enough that a platform fee per minute exceeds an engineer's time amortised across it.
- You already run WebRTC infrastructure and the marginal cost of another workload is near zero.
Outside those, the platform fee is usually cheaper than the engineering, and the honest version of that comparison includes salary rather than stopping at the per-minute rate.