Vapi voice agent latency, independently benchmarked
Vapi is measured in a five-platform benchmark covering Telnyx, Bland AI, ElevenLabs, Retell AI, and Vapi on the same fixed caller script. Time To First Audio Byte is read from the saved audio of real phone calls—not from a platform-reported timestamp.
How Vapi performed
Each platform runs the stack it ships to a new signup, with the documented endpointing exception explained in the complete methodology.
| Metric | Vapi | Standing |
|---|---|---|
| Median TTFAB | 1,558 ms | #4 of 5 |
| p95 TTFAB | 2,008 ms | #3 of 5 |
| Tail ratio (p95 ÷ median) | 1.29× | #2 of 5 |
| Usable turns | 382/432 | 50 discarded |
| Measured invoiced cost/min | $0.0836 | Not a rate-card estimate |
Vapi compared with every measured voice agent platform
The complete table from the main benchmark, ordered by median TTFAB. Read the voice agent latency benchmark for the metric definition, methodology, discard rules, configuration receipts, and open data.
| # | Platform | Official docs | TTFAB p50 | TTFAB p95 | Tail ratio | Usable turns | Cost/min |
|---|---|---|---|---|---|---|---|
| 1 | Telnyx | Telnyx official docs | 1,296 ms | 1,856 ms | 1.43× | 419/432 | $0.0500 |
| 2 | ElevenLabs | ElevenLabs official docs | 1,424 ms | 1,768 ms | 1.24× | 429/432 | $0.0794 |
| 3 | Bland AI | Bland AI official docs | 1,520 ms | 2,248 ms | 1.48× | 429/432 | $0.1408 |
| 4 | Vapi | Vapi official docs | 1,558 ms | 2,008 ms | 1.29× | 382/432 | $0.0836 |
| 5 | Retell AI | Retell AI official docs | 1,740 ms | 2,259 ms | 1.30× | 419/430 | $0.1341 |
Vapi benchmark questions
How fast is Vapi on the voice agent latency benchmark?
Vapi records a median TTFAB of 1,558 ms and a p95 TTFAB of 2,008 ms over 382 usable turns in the current benchmark snapshot. TTFAB is measured from the caller's speech ending to the agent's audio beginning in the saved phone-call recording.
Where are the official Vapi voice agent docs?
The official documentation used to identify the Vapi voice-agent product and setup is linked from this page and the complete benchmark table.
Do any measured voice agent platforms achieve sub-second or under-500 ms latency?
No. None of the 5 measured platforms has a median caller-experienced TTFAB below one second or 500 ms in the current benchmark. The lowest measured median is Telnyx at 1,296 ms. TTFAB here covers the complete pause from the caller finishing speech to hearing the agent's first audio over a real phone call; a server-side vendor TTFB can exclude endpointing, telephony, and network delay.
What are the measured alternatives to Vapi?
The voice agent platforms measured alongside Vapi are Telnyx, ElevenLabs, Bland AI, Retell AI. Every platform is called with the same caller and fixed script, and latency is measured from the saved call audio rather than vendor-reported timestamps.