Historical release. Different cohort and methodology from Benchmark v1; percentages are not directly comparable.
Archived leaderboard
Six platforms · 59 evaluators · 1062 retained calls
| Rank | Platform | pass³ | pass¹ | Median latency | Report |
|---|---|---|---|---|---|
| 01 | Retell | 96.6% | 98.9% | 1.96s | View runs ↗ |
| 02 | Vapi | 94.9% | 98.3% | 2.34s | View runs ↗ |
| 03 | Pipecat | 89.8% | 95.5% | 3.15s | View runs ↗ |
| 04 | LiveKit | 84.7% | 94.9% | 2.46s | View runs ↗ |
| 05 | Synthflow | 81.4% | 90.4% | 3.16s | View runs ↗ |
| 06 | ElevenLabs | 76.3% | 88.1% | 1.73s | View runs ↗ |
v0 attempted to hold one agent and its core model and speech components constant. Latency is the archived median per-turn measure; use this page as historical evidence only.