Cozzo-tested · Round 4 · Reply speed · October 22, 2026
Reply speed: how long AI receptionists make callers wait
A second of silence feels long on the phone. We timed every reply on our receptionist calls, from the moment the caller stopped speaking to the first sound of the agent’s answer, using the call recordings rather than vendor dashboards.
What we found
- ElevenLabs had the fastest median reply at 1.06 s, then Ultravox (1.27 s) and Millis (1.31 s).
- Retell (1.38 s) and OpenAI (1.71 s) sat in the middle. Vapi, Bland and our LiveKit pipeline were all close to 2 seconds.
- Fast isn’t the same as right: ElevenLabs was fastest but passed only 1 of 3 callback read-back calls. Check the other scenarios too.
Look at speed and accuracy together. Our scenario scores and the phone-call latency benchmark measure different things, so use both.
Median reply time by agent
End of caller speech to first agent audio, from the recordings, S1–S3 and S5 calls. Bland dialed in over the phone network, which adds some delay; the others were reached over SIP.
Show full table
| Agent | Median reply | Replies timed |
|---|---|---|
| ElevenLabs Agents | 1.06 s | 20 |
| Ultravox | 1.27 s | 20 |
| Millis AI | 1.31 s | 20 |
| Retell AI | 1.38 s | 20 |
| OpenAI Realtime API | 1.71 s | 30 |
| Vapi | 1.90 s | 20 |
| Bland AI | 1.94 s | 25 |
| LiveKit Agents (self-hosted) | 2.00 s | 27 |
How we timed it
Each reply gap runs from the end of the caller’s speech to the first audio of the agent’s answer, measured on the recording. We used the S1–S3 and S5 calls (S4 has a scripted interruption, so it’s timed separately). Each agent had 20–30 timed replies.
For reply speed on our short booking script, see the voice agent latency benchmark from October 20.
Limits
- 3 calls per scenario per vendor, one dental-office prompt, English only, one evening (Oct 22, 2026). Small differences are noise.
- Read-backs were scored from our AI caller’s speech recognition plus a Whisper transcript of the recording, so a digit misheard on our side is possible. Every call was checked by hand.
- Bland’s scores combine two sessions. Its S4 score mixes 2 scripted cut-ins (both over the limit) with 1 earlier call using our AI caller’s own interruption (passed). Bland dialed in over the phone network; the others were reached over SIP.
- LiveKit is our own self-hosted pipeline (gpt-4o-mini), not a hosted product. OpenAI ran gpt-realtime-2.1 through OpenAI’s SIP endpoint.
- Each hosted agent ran on its default model and voice. Results show what you get before tuning.
Cozzo paid for every call. No vendor sponsored, reviewed or approved these results (see our disclosure).
FAQ
What is a good response time for an AI receptionist?
In our receptionist calls, medians ranged from 1.06 s (ElevenLabs) to 2.00 s (LiveKit). Lower is better: two seconds of silence is a noticeable pause on a phone call.
How did Cozzo measure reply speed?
From the end of the caller’s speech to the first audio of the agent’s reply, on the call recording, across the S1–S3 and S5 calls. Vendor dashboards weren’t used.
Why are these numbers different from the phone-call latency benchmark?
Different test. The October 20 latency benchmark used a short booking script; Round 4 used five longer receptionist scenarios on October 22. Compare vendors within one test, not across tests.
Keep exploring
AI receptionist benchmark
All 5 scenarios and the full scorecard.
All voice AI benchmarks
Every round, plus head-to-head results.
Callback number read-back
We gave 8 AI voice agents the callback number 412-555-0187 on 3 calls each.
Interruption handling
A scripted caller cut in on 8 voice agents.
Emergency calls
We called 8 AI receptionists about an urgent dental problem, 3 times each.
Best AI receptionist for small business
8 receptionist products compared on price and features.
Agents tested: ElevenLabs Agents, Ultravox, Millis AI, Retell AI, OpenAI Realtime API, Vapi, Bland AI, LiveKit Agents (self-hosted).