64.6%
ReliabilityReliability
8th of 15
1.36s
Median replyMedian response time
3rd of 15
$0.0406
Cost / minCost per minute
4th of 11
Higher and further left is better
For the wider field, see our pick of the best voice AI APIs for real-time audio.
Scenarios that passed on all 3 runs, pass³. Higher is better.
PolyAI Dialog-RSN-1 passed 91.0% of booking calls and 8.7% of Medicare calls.
Fields it got wrong most: lead_id, consent_id, route_id
Half of PolyAI Dialog-RSN-1's replies start within 1.36s of the caller finishing, and 19 in 20 within 2.05s. That is 3rd fastest of 15.
Half of replies
P50
9 in 10 replies
P90
19 in 20 replies
P95
Prism
Fastest model · P50
246 live calls: 82 scenarios, each run three times. The median call ran 1:29.
82 scenarios, the same for every model. See every test case
PolyAI Dialog-RSN-1 passed 53 of 82 scenarios on all three runs, 64.6%, 8th of 15 realtime models. Across all 246 calls, 67.9% passed every check.
PolyAI Dialog-RSN-1 passed 53 of 82 scenarios on all three runs, 64.6%, 8th of 15 realtime models. Across all 246 calls, 67.9% passed every check.
PolyAI Dialog-RSN-1 starts replying 1.36s after the caller finishes at the median, 1.87s at P90 and 2.05s at P95. That is 3rd fastest of 15 at the median, measured on the call audio.
When it saved data wrong, the most common mistake was field left out, 78% of them.
PolyAI Dialog-RSN-1 saved everything exactly right on 73.0% of the calls that save data, 11th of 15. On Booking calls, 97.7%. On Medicare calls, 9.1%.
$0.0406 a minute of call at PolyAI's published rates, 4th cheapest of 11 priced models. PolyAI's published rates applied to each call: $0.02 per connected minute plus text tokens ($2.00 per million input, $0.20 cached, $5.00 output). These calls recorded token totals without the cached share, so the share (90% of text input) is taken from four later calls that recorded it; estimated.
Cekura ran PolyAI Dialog-RSN-1 as the whole agent in the same open-source Pipecat pipeline as every other model, with the same prompt, tools and connection. A simulated caller worked through 82 scenarios on live calls, three times each: 246 calls for PolyAI Dialog-RSN-1. A call passes only when the agent said the right things and saved the right data.
Cekura runs these simulated calls on your agent and scores every one.