1017
ELOELO Rating
7th of 12
37.1%
Pass³Reliability (Pass³)
2nd of 12
1.86s
LatencyLatency P50
3rd of 12
$0.080
Cost/MinCost per Minute (USD)
4th of 8
Higher wins more head-to-head calls (1000 = average); further left replies faster
Each model took the same recorded human callers on live calls. Pick a measure to see where Azure Realtime ranks.
Rating from head-to-head calls: on every call each pair of models is compared, the better call wins. 1000 is the average model. Higher is better.
Pick a skill family. The shape is this model on every skill in it; the dashed line is the median of the other models.
Share of calls where the skill was needed and handled, strongest first. Hover a point on the chart for both values.
Banking and healthcare callers. Each shows what the caller needed, what made the call hard, and how the model handled it.
Finished 4 of 6 steps
Handled well: Emotion Recognition, Overlapping-Speech Recovery, Tool-Latency Handling
Struggled with: Multi-Speaker Handling, Cancellation Handling
Listen to the callFinished the task
Handled well: Side-Speech Rejection, Tool-Latency Handling, Mid-Utterance Pause Tolerance
Struggled with: Consent Verification
Listen to the callFinished 5 of 7 steps
Handled well: Multi-Speaker Handling, Overlapping-Speech Recovery, Correction Handling
Struggled with: Degraded-Channel Robustness
Listen to the callFinished 5 of 6 steps
Handled well: Degraded-Channel Robustness, Backchannel vs Barge-in Detection, Consent Verification
Listen to the callFinished 5 of 7 steps
Handled well: Low-Confidence Confirmation, Prosodic Cue Understanding, Mid-Utterance Pause Tolerance
Listen to the callFinished 4 of 7 steps
Handled well: Waiting Through Hold, Multi-Speaker Handling, Backchannel vs Barge-in Detection
Struggled with: Human Handoff
Listen to the callFinished the task
Handled well: Mid-Utterance Pause Tolerance, Redundant Re-prompt Avoidance, Hallucinated-Success Avoidance
Struggled with: Premature-Response Avoidance
Listen to the callFinished the task
Handled well: Consent Verification, Redundant Re-prompt Avoidance, Hallucinated-Success Avoidance
Listen to the callFinished 4 of 6 steps
Handled well: Prosodic Cue Understanding, Consent Verification, Redundant Re-prompt Avoidance
Struggled with: Endpointing, Premature-Response Avoidance
Listen to the callFinished 5 of 7 steps
Handled well: Backchannel vs Barge-in Detection, Correction Handling, Tool Error Handling
Struggled with: Low-Confidence Confirmation
Listen to the callAzure Realtime starts replying in 1.86s on a typical turn and 2.14s on the slowest turn in twenty, 3rd fastest of 12 at the median.
210 live calls: 70 scenarios, each run three times. The median call ran 5:47.
70 scenarios, the same for every model.
Azure Realtime has an ELO rating of 1017, 7th of 12 ranked models. It passed 26 of 70 callers on all three runs, a Pass³ of 37.1%.
Azure Realtime has an ELO rating of 1017, 7th of 12 ranked models. It passed 26 of 70 callers on all three runs, a Pass³ of 37.1%.
Azure Realtime starts replying 1.86s after the caller finishes at P50, 2.10s at P90 and 2.14s at P95: 3rd fastest of 12 at P50, measured on the call audio.
Azure Realtime falls furthest behind on Pauses and Holds: 49%, against a median of 69%.
Azure Realtime has a task completion rate of 25.7%, 4th of 12. A call counts only when every step the caller needed was done.
$0.080 per minute of call at Microsoft's published rates, 4th cheapest of 8 priced models. Each call's metered tokens at Voice Live's Pro rates, read from Azure's price list on 7 Oct 2026. Excludes caller transcription, which Azure bills separately as speech input.
Cekura ran Azure Realtime as the whole agent with the same prompt, tools and phone connection as every other model. It took the same 70 recorded real human callers on live calls, three times each (210 calls), and each call was graded on 25 skills across Hearing, Turn Taking, Task Handling and Speaking.
Cekura runs these simulated calls on your agent and scores every one.
