benchmarks/voice agent latency/telnyx vs retell ai
voice AI latency · head-to-head

Telnyx vs Retell AI — voice agent latency

Two platforms, one identical agent script, dialled over real phone calls: Telnyx currently leads on both readings — the typical pause and the bad one. Every millisecond read from the call's own audio, never from a platform's timestamp.

open data + codeThe caller harness, the offline audio analyzer, and every run's per-turn timings are public in openbenchmarks-labs/voice-agent-latency.github →

Telnyx or Retell AI — measured, reading by reading

The first three rows carry a verdict. The last three do not, and the reason is under the table — a number worth publishing is not always a number worth ranking.

ReadingTelnyxRetell AIMeasured winner
The pause on a normal turn (median TTFAB)1,296 ms1,740 msTelnyx
The pause when it is bad (p95)1,856 ms2,259 msTelnyx
How far the bad turns sit from the normal ones (p95 / median)1.43×1.30×Retell AI
Turns the figures rest on (usable / attempted)419 / 432419 / 430not scored
Turns we refused to time (discarded)1311not scored
Cost per invoiced minute (billing API, pooled)$0.0500$0.1341not scored

The p95 / median column is the two read together — the tail penalty. At 1.2× a platform's bad turns feel close to its normal ones. At 3× a caller regularly waits three times what the median promised.

Why three rows carry no winner. Usable turns is a sample size, not a quality — it says how much the figures above rest on. A discarded turn is one we refused to time, not one the platform failed: the two sides talked over each other, our detectors disagreed, or no reply came, and the largest single reason is our own detectors disagreeing. Ranking platforms on it would score our instrument. Cost per invoiced minute is a real measured charge, but a discounted account tier or a per-call minimum can make two such figures answer different questions, so it is published and left unranked.

route by workflow

There is no single fastest voice agent platform — which one wins depends on which pause your callers actually notice. Pick the workflow that matches yours:

Full ranking, both readings, and the discard counts: the voice agent latency benchmark →

Telnyx vs Retell AI — common questions

Is Telnyx or Retell AI faster for voice agents?

On the pause a caller gets most of the time — median TTFAB — Telnyx is ahead: Telnyx 1,296 ms vs Retell AI 1,740 ms. On the slow turns — p95 — Telnyx is ahead (1,856 ms vs 2,259 ms). Both figures are read from the audio of real phone calls, pooled over usable turns.

What is held equal between Telnyx and Retell AI?

The agent: one system prompt, one greeting, one fixed set of questions in one order, dialled by the same caller over the same carrier. What is deliberately not held equal is each platform's own stack — The stack is not pinned. Each agent runs the model, speech recognition and voice the platform gives a new signup, and we record what it chose. Endpointing is the single exception — it is a timer inside the number we report, so where a platform exposes it we set it to 0.1 s. That one override is stated in full below. Pinning a stack measures a platform you would have to configure to match, not the one you would buy — and it excludes the vertically integrated platforms outright, since you cannot drop a third-party speech recogniser into a platform that owns its own.

Is Telnyx or Retell AI cheaper?

Cost per invoiced minute on this run: Telnyx $0.0500 vs Retell AI $0.1341. No winner is declared on that row, because the two figures are not always the same kind of number — a discounted account tier, a per-call minimum, or an invoice that has not settled for every call all change what "per minute" means. Each platform's qualifiers travel with its figure in cost_notes on the benchmark itself. Read the two as a magnitude, not a ranking.

Why not use the latency Telnyx and Retell AI publish themselves?

A platform measures from where it stands, and the caller is not standing there. We checked that two ways on this bench. A platform's own recording of a call reads roughly 550 ms earlier than our recording of the same call. The latency a platform reports for itself runs roughly 490 ms below what we measure from that call's audio. The two figures agree, and that agreement is the finding: both describe the moment a reply was produced, not the moment a caller heard it. Neither is dishonest — they answer a different question than the one a caller is asking. This board answers the caller's.

What is TTFAB (Time To First Audio Byte)?

Time from the moment the caller stops speaking to the moment the agent's audio starts — the silence a real caller sits through on every turn. Measured from a saved recording of the actual phone call, not from any API timestamp. Lower is better. Also written time to first audio byte, and closely related to what other boards call time to first byte (TTFB) or time to first audio (TTFA).

What does this comparison NOT measure?

Not measured: answer quality, voice quality, platform features, and published pricing plans — this board measures response latency, with the cost each platform actually invoiced for the same run reported beside it. Nor is it every platform. LiveKit Agents and Pipecat are frameworks you host yourself, so what a benchmark would time there is somebody's deployment rather than a product. The raw speech-to-speech APIs — OpenAI's Realtime API, Gemini Live — answer a socket, not a phone, and would need a telephony layer built around them first, which would then be inside the measurement. Twilio's own agent product simply has not been dialled yet. Until any of them is measured on the same script over the same carrier, this board has no number for it, and neither does anyone quoting one. Treat those as vendor claims until measured.