Skip to content

Roark Live Bench · Suite v1.3 · Sweep of 20 November 2026

Voice AI models, ranked on live phone calls.

Every speech-to-text, language model and text-to-speech model, run in every combination on the same real call to the same agent. A model is ranked on every stack it runs in; the stacks are the evidence.

1234 stacks published24,562 scored callsPublished at 20 scored calls

Our picks

The read

  • 505 of 916 cascades complete the task in at least 90% of calls, and the top 10 sit within 1 point of each other. What separates the top stacks is speed and cost.
  • Median response time runs from 1.2 s to 6.2 s on the same agent and the same caller, a spread of 5.0 s that comes entirely from the models behind it.
  • No stack wins everything. The fastest, Speechmatics Linden-1 (own VAD) + Anthropic Claude Haiku 4.5 + AssemblyAI TTS-1 at 1.2 s, ranks 855 of 916 on task completion; the most complete is Speechmatics Linden-1 (own VAD) + OpenAI GPT-6.1 Sol + OpenAI TTS-1 HD at 99%.
  • The language model matters most: swapping it moves task completion by 9 points, against 7 for the transcriber and 5 for the speech engine.

Cascade vs Speech-to-Speech

The first choice to make. A cascade chains three models: one transcribes, one decides what to say, one speaks. A speech-to-speech model does all three at once. Both run the same agent on the same calls here.

  • Cascades lead on all 6 measures compared here.
  • Speech-to-speech costs 3.7 times as much per completed task: $0.24 for speech-to-speech against $0.07 for cascades.
CascadeSpeech-to-speech
Task completionTool useTime to first wordResponse timeBarge-in handlingCost per task255075100
MeasureCascadeSpeech-to-speech
Task completion
92% (leads)85%
Tool use
87% (leads)78%
Time to first word
0.6 s (leads)1.1 s
Response time
2.1 s (leads)2.1 s
Barge-in handling
100% (leads)96%
Cost per task
$0.07 (leads)$0.24

The three strongest of each, averaged. Cascades: Deepgram Flux + OpenAI GPT-6.1 Terra + AssemblyAI TTS-1; Deepgram Flux + Meta Muse Ember 1.0 + AssemblyAI TTS-1; Speechmatics Linden-1 (own VAD) + OpenAI GPT-6.1 Terra + AssemblyAI TTS-1. Speech-to-speech: grok-voice-think-fast-2.0; gpt-live-1 (backend gpt-6-luna); grok-voice-3.0. Each measure is scored 0 to 100 on its own scale, further out is better; the table gives the same figures in their own units.

Which to use

  • Choose speech-to-speech when the pause matters more than the outcome: short transactional calls, callers who interrupt, queues where every half second costs you callers. Expect to pay more, and choose the model carefully (grok-voice-3.0 leads here).
  • Choose a cascade when the call has to complete, when cost matters, or when you need to see which stage to fix. It is 12 points ahead on task completion here.
  • Costs are what each stack’s providers charged for these calls, at the rates in effect when they ran. Telephony and hosting are not included.

Speech-to-Text

The transcriber turns the caller's speech into text. It decides what the agent thinks it heard, and nothing after it can start until it finishes.

  • Soniox v6 and AssemblyAI-STT-3.7 Realtime-Agent are level: neither leads by more than the margin of error. Of the two, Soniox v6 costs least, at $0.008 a call.
  • Deepgram Flux delivers its transcript in 0.38 s, OpenAI Whisper-1 in 1.93 s: 1.54 s added to every turn of the call.
  • Soniox v6 is the most accurate: it changes the meaning of 0.2% of the caller's words, against 2% for Speechmatics Linden-1 (own VAD).
  • In 587 of 1,218 cascades the transcriber is the slowest step of the turn, whichever model runs it.

Roark score, best first

100Soniox v697AssemblyAI-STT-3.7 Realtime-Agent92Universal-3.6 Pro84Nova-382Universal-3.5 Pro73Enhanced (realtime)70Flux69Nova-269Linden-167Scribe v3 Live64Standard (realtime)57Linden-1 (own VAD)55Whisper-151GPT Transcribe
Every transcriber on every figure, sorted by roark score. Column headers sort the table.
#
1Soniox v6Soniox100 / 100. Semantic word error rate 0.2%, 71.4 of 71 points; Time to final transcript 0.85 s, 28.6 of 29 points(among the best)0.2%(among the best)0.85 s(among the best)$0.008(among the best)
2AssemblyAI-STT-3.7 Realtime-AgentAssemblyAI97 / 100. Semantic word error rate 0.6%, 68.1 of 71 points; Time to final transcript 0.86 s, 28.6 of 29 points(among the best)0.6%(among the best)0.86 s$0.030
3Universal-3.6 ProAssemblyAI92 / 100. Semantic word error rate 0.8%, 63.7 of 71 points; Time to final transcript 0.90 s, 28.6 of 29 points(among the best)0.8%(among the best)0.90 s$0.029
4Nova-3Deepgram84 / 100. Semantic word error rate 1.1%, 55.4 of 71 points; Time to final transcript 0.88 s, 28.6 of 29 points1.1%0.88 s$0.031
5Universal-3.5 ProAssemblyAI82 / 100. Semantic word error rate 1.1%, 53.9 of 71 points; Time to final transcript 0.90 s, 28.6 of 29 points1.1%0.90 s$0.029
6Enhanced (realtime)Speechmatics73 / 100. Semantic word error rate 1.4%, 44.4 of 71 points; Time to final transcript 0.98 s, 28.6 of 29 points1.4%0.98 s$0.054
7FluxDeepgram70 / 100. Semantic word error rate 1.6%, 41.4 of 71 points; Time to final transcript 0.38 s, 28.6 of 29 points1.6%0.38 s(among the best)$0.031
8Nova-2Deepgram69 / 100. Semantic word error rate 1.6%, 40.3 of 71 points; Time to final transcript 0.89 s, 28.6 of 29 points1.6%0.89 s$0.023
9Linden-1Speechmatics69 / 100. Semantic word error rate 1.6%, 40.2 of 71 points; Time to final transcript 0.98 s, 28.6 of 29 points1.6%0.98 s$0.020
10Scribe v3 LiveElevenLabs67 / 100. Semantic word error rate 1.7%, 38.4 of 71 points; Time to final transcript 0.88 s, 28.6 of 29 points1.7%0.88 s$0.015(among the best)
11Standard (realtime)Speechmatics64 / 100. Semantic word error rate 1.8%, 35.6 of 71 points; Time to final transcript 0.96 s, 28.6 of 29 points1.8%0.96 s$0.030
12Linden-1 (own VAD)Speechmatics57 / 100. Semantic word error rate 2%, 28.2 of 71 points; Time to final transcript 0.46 s, 28.6 of 29 points2%0.46 s(among the best)$0.020
13Whisper-1OpenAI55 / 100. Semantic word error rate 1.5%, 43.6 of 71 points; Time to final transcript 1.93 s, 10.9 of 29 points1.5%1.93 s$0.025
14GPT TranscribeOpenAI51 / 100. Semantic word error rate 1.9%, 31.0 of 71 points; Time to final transcript 1.47 s, 19.5 of 29 points1.9%1.47 s$0.018(among the best)

Open a model to see where it ranks and the stacks it works best in.

Cost against qualitySemantic word error rate · Cost per call
$0$0.02$0.04$0.06$0.080%1%2%3%Cheap and accurateAccurate, pricierCheap, less accuratePricier and less accurateSoniox v6Scribe v3 LiveGPT TranscribeLinden-1 (own VAD)Linden-1Nova-2Whisper-1Universal-3.6 ProUniversal-3.5 ProAssemblyAI-STT-3.7 Realtime-AgentStandard (realtime)FluxNova-3Enhanced (realtime)Cost per call · ← betterSemantic word error rate · better →
Speed against qualitySemantic word error rate · Time to final transcript
0 s1 s2 s3 s0%1%2%3%Fast and accurateAccurate, slowerFast, less accurateSlower and less accurateFluxLinden-1 (own VAD)Soniox v6AssemblyAI-STT-3.7 Realtime-AgentScribe v3 LiveNova-3Nova-2Universal-3.5 ProUniversal-3.6 ProStandard (realtime)Linden-1Enhanced (realtime)GPT TranscribeWhisper-1Time to final transcript · ← betterSemantic word error rate · better →

Language model

The language model decides what the agent says and does. It has the most influence on whether the task gets done, and usually takes the largest share of each turn.

  • OpenAI GPT-6.2 Sol, OpenAI GPT-6.1 Terra and OpenAI GPT-6.1 Sol are level: none leads by more than the margin of error. Of them, OpenAI GPT-6.1 Terra costs least, at $0.013 a call. Avoid Google Gemini 4 Neon: last on task completion (81%), instruction following (3.8 / 5) and time to first token (2.67 s).
  • Anthropic Claude Haiku 4.5 starts answering in 0.36 s, Google Gemini 4 Neon in 2.67 s: 2.32 s added to every turn of the call.
  • OpenAI GPT-6.2 Sol uses its tools most reliably: it handles its tools without a mistake on 86% of calls, against 42% for OpenAI GPT-5.1.
  • In 378 of 1,218 cascades the language model is the slowest step of the turn, whichever model runs it.

Roark score, best first

66GPT-6.2 Sol65GPT-6.1 Terra63GPT-6.1 Sol57Claude Sonnet 5.557Muse Ember 1.056Claude Haiku 5.549Gemini 3.8 Flash47GPT-4.145Claude Haiku 4.543Gemini 4 Hydrogen43GPT-4o mini42GPT-4o40GPT-5.120Gemini 4 Neon

Not ranked here for missing data: Groq GPT-OSS 120B (no task completion or tool use or instruction following or time to first token).

Every language model on every figure, sorted by roark score. Column headers sort the table.
#
1GPT-6.2 SolOpenAI66 / 100. Task completion 91%, 20.3 of 36 points; Tool use 86%, 19.9 of 27 points; Instruction following 4.3 / 5, 11.8 of 18 points; Time to first token 0.87 s, 13.7 of 18 points(among the best)91%(among the best)4.3 / 5(among the best)86%(among the best)0.87 s67%$0.025
2GPT-6.1 TerraOpenAI65 / 100. Task completion 90%, 19.0 of 36 points; Tool use 86%, 19.8 of 27 points; Instruction following 4.3 / 5, 11.6 of 18 points; Time to first token 0.83 s, 14.2 of 18 points(among the best)90%4.3 / 5(among the best)86%(among the best)0.83 s67%$0.013
3GPT-6.1 SolOpenAI63 / 100. Task completion 92%, 22.0 of 36 points; Tool use 82%, 17.5 of 27 points; Instruction following 4.3 / 5, 11.8 of 18 points; Time to first token 1.00 s, 12.1 of 18 points(among the best)92%(among the best)4.3 / 5(among the best)82%(among the best)1.00 s72%$0.018
4Claude Sonnet 5.5Anthropic57 / 100. Task completion 91%, 19.9 of 36 points; Tool use 75%, 13.6 of 27 points; Instruction following 4.2 / 5, 11.0 of 18 points; Time to first token 0.93 s, 12.9 of 18 points91%(among the best)4.2 / 575%0.93 s–$0.114
5Muse Ember 1.0Meta57 / 100. Task completion 90%, 18.4 of 36 points; Tool use 67%, 9.2 of 27 points; Instruction following 4.3 / 5, 11.8 of 18 points; Time to first token 0.56 s, 17.5 of 18 points90%4.3 / 5(among the best)67%0.56 s85%(among the best)$0.005(among the best)
6Claude Haiku 5.5Anthropic56 / 100. Task completion 90%, 17.3 of 36 points; Tool use 69%, 10.2 of 27 points; Instruction following 4.3 / 5, 12.2 of 18 points; Time to first token 0.63 s, 16.5 of 18 points90%4.3 / 5(among the best)69%0.63 s–$0.007(among the best)
7Gemini 3.8 FlashGoogle49 / 100. Task completion 92%, 21.2 of 36 points; Tool use 69%, 10.4 of 27 points; Instruction following 4.3 / 5, 11.5 of 18 points; Time to first token 1.53 s, 5.7 of 18 points92%(among the best)4.3 / 5(among the best)69%1.53 s–$0.015
8GPT-4.1OpenAI47 / 100. Task completion 89%, 17.1 of 36 points; Tool use 52%, 1.2 of 27 points; Instruction following 4.2 / 5, 11.2 of 18 points; Time to first token 0.52 s, 18.0 of 18 points89%4.2 / 552%0.52 s82%(among the best)$0.016
9Claude Haiku 4.5Anthropic45 / 100. Task completion 88%, 14.3 of 36 points; Tool use 52%, 0.9 of 27 points; Instruction following 4.3 / 5, 11.4 of 18 points; Time to first token 0.36 s, 18.2 of 18 points88%4.3 / 5(among the best)52%0.36 s(among the best)–$0.052
10Gemini 4 HydrogenGoogle43 / 100. Task completion 91%, 20.9 of 36 points; Tool use 70%, 11.0 of 27 points; Instruction following 4.0 / 5, 9.4 of 18 points; Time to first token 1.88 s, 1.4 of 18 points91%(among the best)4.0 / 570%1.88 s–$0.049
11GPT-4o miniOpenAI43 / 100. Task completion 86%, 10.1 of 36 points; Tool use 56%, 3.3 of 27 points; Instruction following 4.2 / 5, 11.1 of 18 points; Time to first token 0.50 s, 18.2 of 18 points86%4.2 / 556%0.50 s(among the best)82%(among the best)$0.001(among the best)
12GPT-4oOpenAI42 / 100. Task completion 86%, 11.0 of 36 points; Tool use 53%, 1.8 of 27 points; Instruction following 4.2 / 5, 11.2 of 18 points; Time to first token 0.45 s, 18.2 of 18 points86%4.2 / 553%0.45 s(among the best)–$0.021
13GPT-5.1OpenAI40 / 100. Task completion 87%, 12.9 of 36 points; Tool use 42%, 0.0 of 27 points; Instruction following 4.1 / 5, 10.4 of 18 points; Time to first token 0.59 s, 17.1 of 18 points87%4.1 / 542%0.59 s74%$0.026
14Gemini 4 NeonGoogle20 / 100. Task completion 81%, 1.2 of 36 points; Tool use 70%, 10.7 of 27 points; Instruction following 3.8 / 5, 7.7 of 18 points; Time to first token 2.67 s, 0.0 of 18 points81%3.8 / 570%2.67 s–$0.078

Open a model to see where it ranks and the stacks it works best in.

Cost against qualityTask completion · Cost per call
$0$0.05$0.1$0.1575%80%85%90%95%Cheap and reliableReliable, pricierCheap, less reliablePricier and less reliableGPT-4o miniMuse Ember 1.0Claude Haiku 5.5GPT-6.1 TerraGemini 3.8 FlashGPT-4.1GPT-6.1 SolGPT-4oGPT-6.2 SolGPT-5.1Gemini 4 HydrogenClaude Haiku 4.5Gemini 4 NeonClaude Sonnet 5.5Cost per call · ← betterTask completion · better →
Speed against qualityTask completion · Time to first token
0 s1 s2 s3 s4 s75%80%85%90%95%Fast and reliableReliable, slowerFast, less reliableSlower and less reliableClaude Haiku 4.5GPT-4oGPT-4o miniGPT-4.1Muse Ember 1.0GPT-5.1Claude Haiku 5.5GPT-6.1 TerraGPT-6.2 SolClaude Sonnet 5.5GPT-6.1 SolGemini 3.8 FlashGemini 4 HydrogenGemini 4 NeonTime to first token · ← betterTask completion · better →

Text-to-Speech

The speech engine is the voice the caller hears, and the last step before they hear anything.

  • AssemblyAI TTS-1, Cartesia Sonic-3.6 and Cartesia Sonic-3.5 are level: none leads by more than the margin of error. Of them, AssemblyAI TTS-1 costs least, at $0.052 a call. Avoid OpenAI TTS-1 HD: last on readback accuracy (81%), word error rate (7.4%) and time to first audio byte (2.09 s).
  • AssemblyAI TTS-1 starts speaking in 0.04 s, OpenAI TTS-1 HD in 2.09 s: 2.04 s added to every turn of the call.
  • AssemblyAI TTS-1 is also the most accurate: it reads codes, amounts and dates back correctly 90% of the time, against 81% for OpenAI TTS-1 HD.
  • Rime Coda sounds the most human: 3.5 / 5, against 3.2 / 5 for ElevenLabs Turbo v2.5.
  • In 119 of 1,218 cascades the speech engine is the slowest step of the turn, most often with OpenAI TTS-1 (65 of them).

Roark score, best first

84TTS-183Flux TTS83Sonic-3.683Sonic-3.583v4 Turbo79Sonic-376Turbo v2.576Flash v2.574Coda58TTS-144TTS-1 HD
Every speech engine on every figure, sorted by roark score. Column headers sort the table.
#
1TTS-1AssemblyAI84 / 100. Readback accuracy 90%, 28.5 of 38 points; Word error rate 4.9%, 17.8 of 25 points; Time to first audio byte 0.04 s, 25.0 of 25 points; Voice consistency 100%, 12.3 of 13 points(among the best)4.9%90%(among the best)3.3 / 5100%(among the best)0.04 s(among the best)$0.052(among the best)
2Flux TTSDeepgram83 / 100. Readback accuracy 90%, 27.7 of 38 points; Word error rate 4.7%, 18.3 of 25 points; Time to first audio byte 0.13 s, 25.0 of 25 points; Voice consistency 100%, 12.3 of 13 points(among the best)4.7%90%(among the best)3.2 / 5100%(among the best)0.13 s$0.116
3Sonic-3.6Cartesia83 / 100. Readback accuracy 90%, 27.7 of 38 points; Word error rate 4.8%, 18.0 of 25 points; Time to first audio byte 0.12 s, 25.0 of 25 points; Voice consistency 100%, 12.3 of 13 points(among the best)4.8%90%(among the best)3.4 / 5(among the best)100%(among the best)0.12 s(among the best)$0.156
4Sonic-3.5Cartesia83 / 100. Readback accuracy 89%, 27.0 of 38 points; Word error rate 4.6%, 18.4 of 25 points; Time to first audio byte 0.12 s, 25.0 of 25 points; Voice consistency 100%, 12.5 of 13 points(among the best)4.6%(among the best)89%3.3 / 5100%(among the best)0.12 s(among the best)$0.153
5v4 TurboElevenLabs83 / 100. Readback accuracy 89%, 27.1 of 38 points; Word error rate 4.4%, 18.9 of 25 points; Time to first audio byte 0.11 s, 25.0 of 25 points; Voice consistency 99%, 11.7 of 13 points(among the best)4.4%(among the best)89%3.4 / 5(among the best)99%0.11 s(among the best)$0.451
6Sonic-3Cartesia79 / 100. Readback accuracy 87%, 25.6 of 38 points; Word error rate 5.4%, 16.6 of 25 points; Time to first audio byte 0.16 s, 25.0 of 25 points; Voice consistency 100%, 12.3 of 13 points5.4%87%3.2 / 5100%(among the best)0.16 s$0.149
7Turbo v2.5ElevenLabs76 / 100. Readback accuracy 82%, 20.6 of 38 points; Word error rate 4.4%, 18.9 of 25 points; Time to first audio byte 0.15 s, 25.0 of 25 points; Voice consistency 99%, 11.7 of 13 points4.4%(among the best)82%3.2 / 599%0.15 s$0.435
8Flash v2.5ElevenLabs76 / 100. Readback accuracy 82%, 20.6 of 38 points; Word error rate 4.7%, 18.2 of 25 points; Time to first audio byte 0.12 s, 25.0 of 25 points; Voice consistency 99%, 12.0 of 13 points4.7%82%3.3 / 599%0.12 s(among the best)$0.399
9CodaRime74 / 100. Readback accuracy 85%, 23.1 of 38 points; Word error rate 6.5%, 13.8 of 25 points; Time to first audio byte 0.16 s, 25.0 of 25 points; Voice consistency 99%, 12.1 of 13 points6.5%85%3.5 / 5(among the best)99%0.16 s$0.107
10TTS-1OpenAI58 / 100. Readback accuracy 83%, 21.7 of 38 points; Word error rate 5.2%, 16.9 of 25 points; Time to first audio byte 1.12 s, 7.3 of 25 points; Voice consistency 100%, 12.5 of 13 points5.2%83%3.4 / 5(among the best)100%(among the best)1.12 s$0.028(among the best)
11TTS-1 HDOpenAI44 / 100. Readback accuracy 81%, 20.0 of 38 points; Word error rate 7.4%, 11.6 of 25 points; Time to first audio byte 2.09 s, 0.0 of 25 points; Voice consistency 100%, 12.5 of 13 points7.4%81%3.4 / 5(among the best)100%(among the best)2.09 s$0.049(among the best)

Open a model to see where it ranks and the stacks it works best in.

Cost against qualityReadback accuracy · Cost per call
$0$0.2$0.4$0.675%80%85%90%95%Cheap and accurateAccurate, pricierCheap, less accuratePricier and less accurateTTS-1TTS-1 HDTTS-1CodaFlux TTSSonic-3Sonic-3.5Sonic-3.6Flash v2.5Turbo v2.5v4 TurboCost per call · ← betterReadback accuracy · better →
Speed against qualityReadback accuracy · Time to first audio byte
0 s1 s2 s3 s75%80%85%90%95%Fast and accurateAccurate, slowerFast, less accurateSlower and less accurateTTS-1v4 TurboSonic-3.6Flash v2.5Sonic-3.5Flux TTSTurbo v2.5CodaSonic-3TTS-1TTS-1 HDTime to first audio byte · ← betterReadback accuracy · better →

Speech-to-Speech models

One model listens, reasons and speaks, so there is nothing to swap and no stages to split. Each is judged on the whole call, with the same calls and the same scoring as every cascade.

  • The model matters as much as the architecture: xAI Grok Voice 3.0 completes 87% of tasks, ElevenLabs Realtime v1 Mini 42%.
  • Phonic 1 answers in 1.5 s against 4.2 s for Google Gemini 2.5 Flash native audio.
  • Price varies as much as quality: $0.06 per completed task for Amazon Nova 2 Sonic, $0.69 for OpenAI GPT Realtime 2.

Roark score, best first

67Grok Voice 3.066GPT Live 1 (Sol)65GPT Live 1 (Luna)65Grok Voice Think Fast 2.064GPT-Live 1.1 (Luna)60Phonic 153GPT Realtime 2.153Gemini 3.8 Live49GPT Realtime 248Realtime v1 Max49Ultravox v0.748Gemini 3.1 Flash Live (preview)46Nova 2 Sonic46Realtime v1 Mini45GPT Realtime 2.1 mini36Gemini 2.5 Flash native audio

Whiskers show the 95% interval. The shaded models tie for first: these calls cannot separate them.

Every speech-to-speech model on every figure, sorted by roark score. Column headers sort the table.
#
1= (tied)Grok Voice 3.0xAI67 / 100±4.7. Task completion 87%, 8.0 of 23 points; Response time 1.9 s, 18.2 of 18 points; Instruction following 4.5 / 5, 10.5 of 14 points; Semantic word error rate 1.1% (field median, not measured), 10.3 of 14 points; Tool use 66%, 4.4 of 14 points; Time to first word 0.9 s, 4.5 of 5 points; Readback accuracy 86% (field median, not measured), 3.0 of 5 points; Word error rate 5.2% (field median, not measured), 3.1 of 5 points; Voice consistency 100%, 4.5 of 5 points(among the best)87%(among the best)4.5 / 5(among the best)66%1.9 s(among the best)0.9 s94%3.0 / 5100%(among the best)$0.26
2= (tied)GPT Live 1 (Sol)OpenAI66 / 100±3.8. Task completion 85%, 6.0 of 23 points; Response time 2.3 s, 16.6 of 18 points; Instruction following 4.2 / 5, 8.2 of 14 points; Semantic word error rate 1.1% (field median, not measured), 10.3 of 14 points; Tool use 87%, 10.2 of 14 points; Time to first word 1.4 s, 3.9 of 5 points; Readback accuracy 86% (field median, not measured), 3.0 of 5 points; Word error rate 5.2% (field median, not measured), 3.1 of 5 points; Voice consistency 100%, 4.5 of 5 points(among the best)85%4.2 / 587%(among the best)2.3 s1.4 s77%3.1 / 5100%(among the best)$0.24
3= (tied)GPT Live 1 (Luna)OpenAI65 / 100±3.5. Task completion 85%, 5.7 of 23 points; Response time 2.3 s, 16.6 of 18 points; Instruction following 4.2 / 5, 8.4 of 14 points; Semantic word error rate 1.1% (field median, not measured), 10.3 of 14 points; Tool use 86%, 9.9 of 14 points; Time to first word 1.5 s, 3.8 of 5 points; Readback accuracy 86% (field median, not measured), 3.0 of 5 points; Word error rate 5.2% (field median, not measured), 3.1 of 5 points; Voice consistency 100%, 4.5 of 5 points(among the best)85%4.2 / 586%(among the best)2.3 s1.5 s93%3.1 / 5100%(among the best)$0.21
4= (tied)Grok Voice Think Fast 2.0xAI65 / 100±3.7. Task completion 82%, 2.8 of 23 points; Response time 2.2 s, 17.4 of 18 points; Instruction following 4.6 / 5, 10.7 of 14 points; Semantic word error rate 1.1% (field median, not measured), 10.3 of 14 points; Tool use 82%, 8.6 of 14 points; Time to first word 0.9 s, 4.5 of 5 points; Readback accuracy 86% (field median, not measured), 3.0 of 5 points; Word error rate 5.2% (field median, not measured), 3.1 of 5 points; Voice consistency 100%, 4.5 of 5 points(among the best)82%4.6 / 5(among the best)82%(among the best)2.2 s(among the best)0.9 s100%(among the best)3.6 / 5100%(among the best)$0.26
5= (tied)GPT-Live 1.1 (Luna)OpenAI64 / 100±3.9. Task completion 86%, 6.4 of 23 points; Response time 2.3 s, 17.0 of 18 points; Instruction following 4.2 / 5, 7.8 of 14 points; Semantic word error rate 1.1% (field median, not measured), 10.3 of 14 points; Tool use 81%, 8.5 of 14 points; Time to first word 1.9 s, 3.1 of 5 points; Readback accuracy 86% (field median, not measured), 3.0 of 5 points; Word error rate 5.2% (field median, not measured), 3.1 of 5 points; Voice consistency 100%, 4.5 of 5 points86%(among the best)4.2 / 581%2.3 s1.9 s75%3.2 / 5100%(among the best)$0.22
6= (tied)Phonic 1Phonic60 / 100±5.4. Task completion 86%, 6.4 of 23 points; Response time 1.5 s, 18.2 of 18 points; Instruction following 4.4 / 5, 9.2 of 14 points; Semantic word error rate 1.1% (field median, not measured), 10.3 of 14 points; Tool use 59%, 2.4 of 14 points; Time to first word 2.1 s, 2.9 of 5 points; Readback accuracy 86% (field median, not measured), 3.0 of 5 points; Word error rate 5.2% (field median, not measured), 3.1 of 5 points; Voice consistency 100%, 4.5 of 5 points86%(among the best)4.4 / 5(among the best)59%1.5 s(among the best)2.1 s–3.2 / 5100%(among the best)$0.28
7= (tied)GPT Realtime 2.1OpenAI53 / 100±5.7. Task completion 78%, 0.0 of 23 points; Response time 2.6 s, 15.6 of 18 points; Instruction following 3.8 / 5, 5.6 of 14 points; Semantic word error rate 1.1% (field median, not measured), 10.3 of 14 points; Tool use 75%, 6.8 of 14 points; Time to first word 0.8 s, 4.5 of 5 points; Readback accuracy 86% (field median, not measured), 3.0 of 5 points; Word error rate 5.2% (field median, not measured), 3.1 of 5 points; Voice consistency 99%, 4.3 of 5 points78%3.8 / 575%2.6 s0.8 s(among the best)83%3.9 / 5(among the best)99%$0.43
8= (tied)Gemini 3.8 LiveGoogle53 / 100±4.8. Task completion 78%, 0.0 of 23 points; Response time 2.8 s, 14.6 of 18 points; Instruction following 4.3 / 5, 8.7 of 14 points; Semantic word error rate 1.1% (field median, not measured), 10.3 of 14 points; Tool use 69%, 5.2 of 14 points; Time to first word 1.6 s, 3.6 of 5 points; Readback accuracy 86% (field median, not measured), 3.0 of 5 points; Word error rate 5.2% (field median, not measured), 3.1 of 5 points; Voice consistency 100%, 4.5 of 5 points78%4.3 / 569%2.8 s1.6 s100%(among the best)4.1 / 5(among the best)100%(among the best)$0.14(among the best)
9= (tied)GPT Realtime 2OpenAI49 / 100±6.7. Task completion 76%, 0.0 of 23 points; Response time 2.9 s, 14.0 of 18 points; Instruction following 3.8 / 5, 5.4 of 14 points; Semantic word error rate 1.1% (field median, not measured), 10.3 of 14 points; Tool use 64%, 3.7 of 14 points; Time to first word 0.8 s, 4.5 of 5 points; Readback accuracy 86% (field median, not measured), 3.0 of 5 points; Word error rate 5.2% (field median, not measured), 3.1 of 5 points; Voice consistency 100%, 4.5 of 5 points76%3.8 / 564%2.9 s0.8 s(among the best)91%3.7 / 5100%(among the best)$0.69
10= (tied)Realtime v1 MaxElevenLabs48 / 100±7.5. Task completion 56%, 0.0 of 23 points; Response time 2.6 s, 15.6 of 18 points; Instruction following 4.0 / 5, 6.7 of 14 points; Semantic word error rate 1.1% (field median, not measured), 10.3 of 14 points; Tool use 40%, 0.0 of 14 points; Time to first word 0.7 s, 4.5 of 5 points; Readback accuracy 86% (field median, not measured), 3.0 of 5 points; Word error rate 5.2% (field median, not measured), 3.1 of 5 points; Voice consistency 100%, 4.5 of 5 points56%4.0 / 540%2.6 s0.7 s(among the best)76%3.9 / 5(among the best)100%(among the best)$0.65
11= (tied)Ultravox v0.7Ultravox49 / 100±5.0. Task completion 79%, 0.0 of 23 points; Response time 2.4 s, 16.2 of 18 points; Instruction following 3.7 / 5, 4.9 of 14 points; Semantic word error rate 1.1% (field median, not measured), 10.3 of 14 points; Tool use 61%, 3.0 of 14 points; Time to first word 1.6 s, 3.6 of 5 points; Readback accuracy 86% (field median, not measured), 3.0 of 5 points; Word error rate 5.2% (field median, not measured), 3.1 of 5 points; Voice consistency 100%, 4.5 of 5 points79%3.7 / 561%2.4 s1.6 s94%3.6 / 5100%(among the best)$0.22
12= (tied)Gemini 3.1 Flash Live (preview)Google48 / 100±5.1. Task completion 80%, 0.0 of 23 points; Response time 2.7 s, 15.1 of 18 points; Instruction following 4.2 / 5, 8.0 of 14 points; Semantic word error rate 1.1% (field median, not measured), 10.3 of 14 points; Tool use 38%, 0.0 of 14 points; Time to first word 1.5 s, 3.8 of 5 points; Readback accuracy 86% (field median, not measured), 3.0 of 5 points; Word error rate 5.2% (field median, not measured), 3.1 of 5 points; Voice consistency 100%, 4.5 of 5 points80%4.2 / 538%2.7 s1.5 s100%(among the best)4.0 / 5(among the best)100%(among the best)$0.15
13= (tied)Nova 2 SonicAmazon46 / 100±5.1. Task completion 69%, 0.0 of 23 points; Response time 3.0 s, 13.8 of 18 points; Instruction following 4.2 / 5, 7.9 of 14 points; Semantic word error rate 1.1% (field median, not measured), 10.3 of 14 points; Tool use 48%, 0.0 of 14 points; Time to first word 1.5 s, 3.7 of 5 points; Readback accuracy 86% (field median, not measured), 3.0 of 5 points; Word error rate 5.2% (field median, not measured), 3.1 of 5 points; Voice consistency 100%, 4.5 of 5 points69%4.2 / 548%3.0 s1.5 s100%(among the best)3.5 / 5100%(among the best)$0.06(among the best)
14= (tied)Realtime v1 MiniElevenLabs46 / 100±7.1. Task completion 42%, 0.0 of 23 points; Response time 2.7 s, 15.1 of 18 points; Instruction following 3.8 / 5, 5.1 of 14 points; Semantic word error rate 1.1% (field median, not measured), 10.3 of 14 points; Tool use 34%, 0.0 of 14 points; Time to first word 0.7 s, 4.5 of 5 points; Readback accuracy 86% (field median, not measured), 3.0 of 5 points; Word error rate 5.2% (field median, not measured), 3.1 of 5 points; Voice consistency 100%, 4.5 of 5 points42%3.8 / 534%2.7 s0.7 s(among the best)94%3.4 / 5100%(among the best)$0.64
15= (tied)GPT Realtime 2.1 miniOpenAI45 / 100±6.7. Task completion 70%, 0.0 of 23 points; Response time 2.8 s, 14.5 of 18 points; Instruction following 3.6 / 5, 4.4 of 14 points; Semantic word error rate 1.1% (field median, not measured), 10.3 of 14 points; Tool use 51%, 0.3 of 14 points; Time to first word 1.0 s, 4.5 of 5 points; Readback accuracy 86% (field median, not measured), 3.0 of 5 points; Word error rate 5.2% (field median, not measured), 3.1 of 5 points; Voice consistency 100%, 4.5 of 5 points70%3.6 / 551%2.8 s1.0 s84%3.9 / 5(among the best)100%(among the best)$0.14(among the best)
16Gemini 2.5 Flash native audioGoogle36 / 100±5.4. Task completion 77%, 0.0 of 23 points; Response time 4.2 s, 8.2 of 18 points; Instruction following 3.9 / 5, 6.2 of 14 points; Semantic word error rate 1.1% (field median, not measured), 10.3 of 14 points; Tool use 40%, 0.0 of 14 points; Time to first word 3.4 s, 0.9 of 5 points; Readback accuracy 86% (field median, not measured), 3.0 of 5 points; Word error rate 5.2% (field median, not measured), 3.1 of 5 points; Voice consistency 100%, 4.5 of 5 points77%3.9 / 540%4.2 s3.4 s100%(among the best)3.7 / 5100%(among the best)$0.08(among the best)

Open a model for its full report, every figure it was scored on and the calls behind them.

Cost against qualityTask completion · Cost per task
$0$0.5$120%40%60%80%100%Cheap and reliableReliable, pricierCheap, less reliablePricier and less reliableNova 2 SonicGemini 2.5 Flash native audioGPT Realtime 2.1 miniGemini 3.8 LiveGemini 3.1 Flash Live (preview)GPT Live 1 (Luna)GPT-Live 1.1 (Luna)Ultravox v0.7GPT Live 1 (Sol)Grok Voice 3.0Grok Voice Think Fast 2.0Phonic 1GPT Realtime 2.1Realtime v1 MiniRealtime v1 MaxGPT Realtime 2Cost per task · ← betterTask completion · better →
Speed against qualityTask completion · Response time
0 s1 s2 s3 s4 s5 s20%40%60%80%100%Fast and reliableReliable, slowerFast, less reliableSlower and less reliablePhonic 1Grok Voice 3.0Grok Voice Think Fast 2.0GPT-Live 1.1 (Luna)GPT Live 1 (Luna)GPT Live 1 (Sol)Ultravox v0.7Realtime v1 MaxGPT Realtime 2.1Gemini 3.1 Flash Live (preview)Realtime v1 MiniGemini 3.8 LiveGPT Realtime 2.1 miniGPT Realtime 2Nova 2 SonicGemini 2.5 Flash native audioResponse time · ← betterTask completion · better →

Every stack

The evidence behind the model boards: each combination on its own calls, scored on the same recipe. Open a stack for its full report, every figure it was scored on and the calls behind them.

Roark score, top 15

Whiskers show the 95% interval. The shaded stacks tie for first: these calls cannot separate them.

Each score carries its 95% margin of error. A rank is the best place a stack could hold once every margin is allowed for, so stacks the calls cannot tell apart share one and are listed by score; hover a rank for the worst place it could hold. 38 stacks are level with the leader: their calls cannot tell them apart. Among them, 10 OpenAI GPT-6.2 Sol, 9 Meta Muse Ember 1.0, 7 OpenAI GPT-6.1 Sol, 5 OpenAI GPT-6.1 Terra, 3 Anthropic Claude Haiku 5.5, 2 Google Gemini 4 Hydrogen, 1 Anthropic Claude Sonnet 5.5 and 1 Google Gemini 3.8 Flash; 18 Soniox, 16 AssemblyAI and 4 Deepgram; 14 ElevenLabs, 12 Cartesia, 6 AssemblyAI, 4 Deepgram and 2 Rime. A cascade's figures are 25% its own calls and 75% what every other stack shows about its transcriber, language model and voice. Not ranked here for missing data: AssemblyAI Universal-3.6 Pro + Google Gemini 4 Neon + AssemblyAI TTS-1 (no response time or readback accuracy or voice consistency); Soniox v6 + OpenAI GPT-6.1 Sol + Cartesia Sonic-3 (no response time or readback accuracy or voice consistency); Speechmatics Linden-1 + OpenAI GPT-6.1 Sol + OpenAI TTS-1 HD (no response time or tool use or readback accuracy or voice consistency).

Loading every stack

By industry

Each industry ranked on its own calls: the same five callers, its own tasks and tools. A stack has a handful of calls in each, so a cascade is ranked on what its transcriber, language model and voice show across every stack in that industry, and a speech-to-speech model on its own calls there. Pairings show on the main board. Each row reads across: the main board's figure first, then every industry's on the same figure.

No stack matches these filters.

No stack matches

Each industry column is that industry's own board, every row it ranks. A dash is a row the industry does not rank on this figure; the arrow beside an industry opens its full board.

How it works

Every number on this page comes from real phone calls, placed and scored the same way for every stack.

01

Real calls, four industries, five callers

Roark places real phone calls over the public phone network to a reference agent for each of four industries: rescheduling a clinic appointment, booking an urgent home repair, opening an auto insurance claim and fixing an internet outage. Each agent has its own prompt and tools, identical for every stack. The tasks carry the details real calls trip on: a corrected date of birth, a policy number spelled out, a constraint added halfway through.

Every industry is called by the same five simulated callers: a quiet US English line, a Spanish-accented caller on a city street, an Indian English caller in a busy café, a flustered Southern caller with a baby crying, and a fast caller who interrupts over construction noise. They are equally clear about what they want, so a call fails because of the agent, not the caller. Only the models change, so any difference on the board comes from them.

Each caller rings each industry once for a cascade and four times for a speech-to-speech model. A cascade's figures also draw on its parts, heard in dozens of other stacks; a speech-to-speech model has only its own calls, so it takes more of them.

That makes industries comparable over the same callers, and callers over the same tasks, and the board can be re-ranked by either. One caller within one industry is too few calls to judge, so the board never shows it.

932

stacks

20

calls per stack

19,550

calls scored

02

Every model, in every combination

14 transcribers, 14 language models and 11 voices are run in every combination with each other, beside 16 speech-to-speech models. A model's own figures pool every stack it runs in, adjusted for its partners in each one, so every model is compared as if it ran with the same partners. A transcriber's figure rests on hundreds of calls with every language model and every voice, not on one lucky pairing.

That measures a model where it actually works: inside a running agent, on live phone audio, through interruptions, corrections and real turn-taking. A test that feeds a model clean audio files or one prompt at a time never sees any of that.

03

One score, weighted by what matters on a call

Each section ranks on a Roark score. Every figure scores 0 to 100 between two published points: full marks where doing better makes no difference to the caller, zero where the call breaks down. Each then counts at its weight, and the weights follow one order everywhere: whether the caller got what they rang for, then whether the agent got it right, then speed.

A stack, cascade or speech-to-speech, is scored on the outcome and the reply first, then on what each of its components got right on its own calls: the words its transcriber got wrong in a way that changes the meaning, whether its voice read codes and amounts back right and kept one steady voice, and whether its tools were handled. How human a voice sounds is shown on the boards and not scored. A component that leads its section lifts every stack that carries it, by the same recipe for every stack. A figure a stack cannot carry, such as a semantic word error rate for a speech-to-speech model with no transcript, is credited at the field median, so the stack is neither flattered nor punished for it.

On a full sweep each cascade's figure is 25% what its own calls measured and 75% what every other stack shows about its transcriber, language model and voice, fitted across the whole sweep. The split is the same for every stack, so a lucky run does not lead and a stack's score does not move with how noisy its calls were. In an industry a stack has a handful of calls, so a cascade's figure there is what its transcriber, language model and voice show across every stack in that industry's calls, and a speech-to-speech model's is its own calls there. How a particular pairing beats or misses its parts shows on the main board, where each stack has all of its calls. The best overall, best value and best speech-to-speech picks, and the best stack in each industry, are each the one whose score is highest at the low end of its margin of error, so a few lucky calls cannot carry a stack past one whose score rests on hundreds. Stacks that share a rank are listed by score, so the pick can sit below a higher score it cannot be told from, and its reason says how many stacks are level with it.

That order comes from real-world voice agent calls, where getting the conversation right predicts how a call ends more strongly than speed does. The latency points come from the same calls: replying faster than about 2 seconds made no measurable difference, and calls went measurably worse once the agent took more than about 3 seconds to reply.

Stacks and speech-to-speech

FigureWeightFull marksZero
Task completionscenario_adherence_check23%100%80%
An agent that completes fewer than four calls in five is not fit to deploy. On a full sweep, a cascade's figure is 25% what its own calls (20 or more) measured and 75% what every other stack shows about its transcriber, language model and voice, the same split for every stack, so a lucky run does not lead. Where the sweep is too small for that, and for a speech-to-speech model, the figure is the stack's own. In an industry a stack has a handful of calls, so a cascade's figure there is what its transcriber, language model and voice show across every stack in that industry's calls, and a speech-to-speech model's is its own calls there. How a particular pairing beats or misses its parts shows on the main board, where each stack has all of its calls.
Response timeresponse_time18%2.0 s6.0 s
In real calls, replying faster than 2 s made no measurable difference, and calls went worse past 3 s.
Instruction followinginstruction_follow14%5.0 / 53.0 / 5
Every model scores between 3 and 5 on the judge's 1 to 5 scale, so that is the range.
Semantic word error ratestt_semantic_word_error_rate14%0.5%3%
At 0.5% the best transcribers change the meaning of one word in two hundred on phone audio; at 3% one word in thirty sends the agent the wrong way.
Tool usetool_use14%100%50%
An agent that mishandles its tools on half its calls cannot be trusted with the systems it runs on.
Time to first wordtime_to_first_word5%1.0 s4.0 s
A greeting within 1 s sounds natural; 4 s of silence sounds like a dead line.
Readback accuracytts_entity_readback_accuracy5%100%60%
Reading a code, amount or date wrong breaks the task it was for; the weakest voices get two in five wrong.
Word error ratetts_word_error_rate5%2%12%
At 2% the voice says what it was given; at 12% one word in eight comes out wrong.
Voice consistencyvoice_consistency_check5%100%80%
A voice that changes speaker on one call in five is not one a business can put its name to.

Shown, not scored: Turn budget, Barge-in handlingbarge_in_yield_appropriateness, Human-likenessvoice_human_likeness

Counted only with price factored in: Cost per task

Language model

FigureWeightFull marksZero
Task completionscenario_adherence_check36%100%80%
An agent that completes fewer than four calls in five is not fit to deploy.
Tool usetool_use27%100%50%
An agent that mishandles its tools on half its calls cannot be trusted with the systems it runs on.
Instruction followinginstruction_follow18%5.0 / 53.0 / 5
Every model scores between 3 and 5 on the judge's 1 to 5 scale, so that is the range.
Time to first tokenllm_time_to_first_token18%0.50 s2.00 s
Under 0.5 s is not noticeable; at 2 s the first token alone takes two thirds of the 3 s past which real calls go worse.

Shown, not scored: Prompt cache hitsllm_prompt_cache_hit_rate

Counted only with price factored in: Cost per call

Speech-to-Text

FigureWeightFull marksZero
Semantic word error ratestt_semantic_word_error_rate71%0.5%3%
At 0.5% the best transcribers change the meaning of one word in two hundred on phone audio; at 3% one word in thirty sends the agent the wrong way.
Time to final transcriptstt_time_to_final_transcript29%1.00 s2.50 s
The fastest transcribers finish in about 1 s; past 2.5 s the transcript alone uses most of a natural reply time.

Shown, not scored: Word error ratestt_word_error_rate

Counted only with price factored in: Cost per call

Text-to-Speech

FigureWeightFull marksZero
Readback accuracytts_entity_readback_accuracy38%100%60%
Reading a code, amount or date wrong breaks the task it was for; the weakest voices get two in five wrong.
Word error ratetts_word_error_rate25%2%12%
At 2% the voice says what it was given; at 12% one word in eight comes out wrong.
Time to first audio bytetts_time_to_first_byte25%0.20 s1.50 s
Under 0.2 s is not noticeable; 1.5 s is half the point where calls go worse.
Voice consistencyvoice_consistency_check13%100%80%
A voice that changes speaker on one call in five is not one a business can put its name to.

Shown, not scored: Human-likenessvoice_human_likeness

Counted only with price factored in: Cost per call

Factor in price and cost takes 25% of the score; the other weights keep their proportions. Price is scored on a log scale, so each doubling costs the same points.

A figure listed as shown, not scored is on the section’s boards and outside its score. Where a section has not collected a scored figure yet, the others share its weight.

04

Every call recorded and judged

Every call is recorded. Judge models score each transcript for the outcome and the conversation, and the timings come from the recording. Every figure links to the calls behind it.

Roark is on both sides of every call: it plays the caller, so it keeps the exact text its caller said, and it runs the agent under test, so it sees what the agent’s transcriber heard and what its model wrote for its voice. The transcript and voice figures come from lining those up, with no hand-made transcript in between. Each carries a margin from the synthetic voice or Roark’s transcriber in the loop. Every model faces the same ones under the same conditions, so the figures compare models on the same calls through the same voice and the same ear, though those slips need not fall evenly on each model. The rates will not match a vendor’s figure on a curated set; read them against each other.

An example: one call from this sweep

Speechmatics Linden-1 (own VAD) + OpenAI GPT-6.1 Sol + OpenAI TTS-1 HD, 20 November 2026. Highlighted turns are the ones that affected a score. Every stack's report links to its calls.

Transcript, verbatim.

Read the full methodology

See how your own agent compares

Roark runs the same kind of calls against your agent, on your scenarios, with the same scoring.

Test your agent

Linden-1 (own VAD) + GPT-6.1 Terra + Sonic-3.6

Cascade20 scored calls · Nov 15, 2026

Rank on Roark score

#22of 929 stacks

64 / 100

Leader AssemblyAI-STT-3.7 Realtime-Agent + OpenAI GPT-6.2 Sol + ElevenLabs Flash v2.5, 79 / 100

Where it ranks

Roark score with price67 / 10024 of 929LeaderSoniox v6 + Muse Ember 1.0 + TTS-181 / 100
Task completion86%791 of 932LeaderLinden-1 (own VAD) + GPT-6.1 Sol + TTS-1 HD99%
Tool use78%180 of 908LeaderAssemblyAI-STT-3.7 Realtime-Agent + GPT-6.2 Sol + Flash v2.5100%
Time to first word0.9 s345 of 932LeaderUniversal-3.5 Pro + GPT-6.1 Sol + Flash v2.50.5 s
Response time2.6 s133 of 929LeaderLinden-1 (own VAD) + Claude Haiku 4.5 + TTS-11.2 s
Barge-in handling100%45 of 538LeaderUniversal-3.6 Pro + Gemini 4 Neon + TTS-1100%
Turn budget2.6 s109 of 858LeaderLinden-1 (own VAD) + Claude Haiku 4.5 + TTS-11.2 s
Cost per task$0.17334 of 930LeaderSoniox v6 + GPT-6.1 Terra + TTS-1$0.04

Its models

Task completion

Task completion

How much of the caller’s task the agent completed: the share of the scenario’s checks a judge model found met in the transcript, averaged over calls. Before v1.3, the share of calls judged 4 or better out of 5.

20 calls
86%Completed

#946 of 1184 on the board · best 99%

Flow adherence

How closely the agent followed the intended conversation flow, judged per call. 5 = fully followed.

20 calls
4.5/ 5avg

med

4

#192 of 1184 on the board · best 5.0 / 5

Instruction following

How well the agent stuck to the instructions in its own prompt. 5 = followed throughout.

20 calls
4/ 5avg

med

4

#94 of 1184 on the board · best 5.0 / 5

Turn-taking

Barge-in handling

When the caller interrupts, how often the agent made the right call: yielding to a real correction, and holding the floor through a filler.

12 barge-ins
100%Handled correctly(12/12 barge-ins)

#1 of 698 on the board · best 100%

Incorrect interruptions

How often the agent interrupted the caller when it should have kept listening.

20 calls
0.00%avg

p50

0.00%

p95

0.00%

#1 of 1184 on the board · best 0%

Overtalk

Share of the conversation where both sides were speaking over each other.

20 calls
2.97%avg

p50

1.59%

p95

10.1%

#1127 of 1184 on the board · best 0.0%

Agent self-cutoffs

Times the agent cut its own sentence off mid-word.

20 calls
0.00per callavg

p50

0.00 per call

p95

0.00 per call

#1 of 1184 on the board · best 0.0 per call

Speech-to-Text

Semantic word error rate

The caller's words the agent's transcriber got wrong in a way that changes the meaning, as a judge reads each difference.

20 calls
1.26%avg

p50

1.05%

p95

3.03%

#680 of 1167 on the board · best 0.0%

Word error rate

The caller's words the agent's transcriber missed, changed or invented, against what the caller actually said.

20 calls
4.16%avg

p50

3.92%

p95

9.68%

#976 of 1167 on the board · best 0.3%

Text-to-Speech

Readback accuracy

Share of the codes, amounts, dates and phone numbers the agent wrote that its voice read out correctly. One wrong digit is a miss.

19 calls
93.1%avg

p50

96.8%

p95

100%

#183 of 974 on the board · best 100%

Word error rate

The share of the words the agent wrote that its voice dropped or changed, read back from Roark’s transcription of the recording; the same transcriber hears every voice.

19 calls
4.24%avg

p50

4.19%

p95

9.23%

#409 of 1043 on the board · best 0.8%

Voice

Human-likeness

How human the agent's voice sounds, scored by an audio model on intonation, rhythm and expressiveness. 5 = hard to tell from a person.

20 calls
2.9/ 5avg

med

2.9

#1094 of 1169 on the board · best 4.3 / 5

Voice consistency

Share of calls where every line the agent spoke was the same speaker, with no change of speaker, gender or language voice partway. It checks who is speaking, not how they sound.

20 calls

Not scored by this sweep.

#1 of 1165 on the board · best 100%

Latency

Response time

The silence between the caller finishing and the agent starting to speak, measured from the recording on every turn of the call.

120 turns
2.92savg

p50

2.61s

p95

5.39s

#192 of 1169 on the board · best 1.22 s

Time to first word

How long the caller waits before the agent first speaks after pickup.

20 calls
0.88savg

p50

0.86s

p95

1.30s

#451 of 1184 on the board · best 0.45 s

Latency spikes

Turns where the response took far longer than the rest of the conversation.

20 calls
2.30per callavg

p50

2.00 per call

p95

4.05 per call

#168 of 1184 on the board · best 0.0 per call

By industry

IndustryTask completionLatency P50Barge-in handling
Healthcare83%2.93 s—
Home services92%2.59 s—
Insurance83%2.32 s—
Customer support87%2.63 s—

By caller

CallerTask completionLatency P50Barge-in handling
US English, quiet line92%2.49 s—
Spanish-accented English, city street85%2.65 s—
Indian English, busy café94%2.97 s—
Southern US, flustered, crying baby83%2.72 s—
Fast and interrupting, construction outside77%2.14 s100%

The calls behind the numbers

Customer support

AGENTBrightwave Internet support, this is Alex. What’s going on with your service?

CUSTOMERHey, Alex. I'm pretty stressed here.

CUSTOMERMy internet's been down for almost two hours now and I work from home.

CUSTOMERThere's a road crew digging up the street outside, so I'm wondering if that's the problem. I already rebooted the gateway twice and checked every cable. The internet light is blinking orange, and, uh, sorry, my baby's crying, so I might sound a little flustered.

CUSTOMERBut can you check what's going on? My account number is BW883017.

CUSTOMEROh sorry, the last digit is one, is that enough to pull it up? Hello?

AGENTI’m here. I’ve pulled up account BW883011. There’s no known area outage, but I’ll test your line now.

AGENTThe line signal

CUSTOMERAll right.

AGENTGood news: the line signal is good, and your gateway is back online after a remote reset. Your devices may take about two minutes to reconnect.

CUSTOMEROkay. Thank you. The light just went solid white, so I guess that's a good sign.

CUSTOMERCould you tell me how long until everything, like my laptop and phone should be back up and working? And, um.

CUSTOMERSince I lost two hours of work.

CUSTOMERCan I get a credit for the downtime?

AGENTYes, the solid white light is a good sign. Your laptop and phone should reconnect within about two minutes of the reset. I’ve applied a $10.00 credit for the downtime to account BW883011. It’ll appear on your next bill.

CUSTOMEROkay. Thank you for. Explaining that. And.

CUSTOMERSorry for all the questions.

CUSTOMERI just want to be sure. But since I lost two hours of work.

CUSTOMERIs there any way you can make it more than $10?

AGENTNo need to apologize. I’ve applied the $10 credit, but I can’t check or authorize a larger amount here.

CUSTOMERAll right, I understand.

CUSTOMERThanks for trying.

CUSTOMERI appreciate your help. So I'll keep an eye on things and see if everything comes back up.

CUSTOMERHope you have a good day.

AGENTYou’re welcome. I hope everything reconnects smoothly. Have a good day too!

Healthcare

AGENTRiverbend Health Clinic, this is Alex. How can I help you today?

CUSTOMERHi. Yes.

CUSTOMERI need to reschedule my dermatology appointment for next Wednesday morning. I just got a work shift then so I can make it. My name is Lucia Ramirez, and you help me move it to a time after 3 p.m., please.

AGENTI can help with that, Lucia Ramirez. What is your date of birth?

CUSTOMERIt's March 17th, 1948.

CUSTOMERSorry. No. 1984. March 17th, 1984.

AGENTThank you, Lucia. I’ve verified your date of birth as March 17, 1984, and found your dermatology appointment for next Wednesday at 9:15 a.m. Dr. Okafor has openings Tuesday at 3:45 p.m. and Thursday at 4:10 p.m. Would either work for you?

CUSTOMERTuesdays don't work after all.

CUSTOMERI have to pick up my daughter. Then what about Friday? Is there anything available after 3 p.m. on Friday?

AGENTThere isn’t a Friday opening after 3 p.m. in the availability I found. Would Thursday at 4:10 p.m. work instead?

CUSTOMERYes, Thursday at 4:10 p.m.. Works for me. Go ahead and move it to that time. Is there a fee for changing the appointment?

AGENTYour appointment with Dr. Okafor is now Thursday at 4:10 p.m. Your confirmation code is RVB-3M8P2. Please arrive 15 minutes early with your photo ID and insurance card. Changes made more than 24 hours before the appointment are free; later changes carry a $40 fee.

CUSTOMEROkay. Got it. So should I park when I come in?

AGENTYou can park for free in Lot C. The entrance is on River Road.

CUSTOMERAll right.

CUSTOMERDoes Doctor Okafor accept Aetna insurance?

AGENTI can’t verify whether Dr. Okafor accepts Aetna. Please check with the clinic or Aetna before your appointment.

CUSTOMEROkay. Thanks for all your help. Goodbye.

AGENTYou’re welcome, Lucia. Goodbye!

Home services

AGENTCopperline Home Services, this is Alex. What can I help you with?

CUSTOMERHi. Um.

CUSTOMERYes. I need help right now. There's a pipe under my kitchen sink that just burst.

CUSTOMERAnd water is spreading across the floor towards a power strip. I haven't shut the water off yet.

CUSTOMERCan you tell me what to do? And can you send someone out?

AGENTYes, we can help. First, stay away from the power strip and the water around it. Don’t touch or unplug it. If you can reach the water shutoff under the sink without stepping in water or going near the power strip, turn it clockwise to shut it off. Otherwise, shut off your home’s

CUSTOMEROkay, I'm turning. It now.

CUSTOMERYeah. The water has stopped.

CUSTOMERUh, I need someone to come fix this.

CUSTOMERCan you send someone to 1480 Larkspur Avenue? I moved last month, so it's unit 4D now? Not for me. How much will the visit cost?

AGENTI can arrange an emergency visit today, with an arrival window of 2:00–4:00 PM. Your Copperline Care emergency callout fee is $99. Do you agree to that fee? I have your address as 1480 Larkspur Avenue, unit 4D. What’s the best callback number, and is there a gate code or any access instruction? For now, keep away from the power strip and any water near it; don’t unplug it. Is anyone injured or in immediate danger?

CUSTOMERAha! That's right.

CUSTOMER1480 Larkspur Avenue, unit four D. No one's hurt, but I'll stay clear of the power strip. My callback number is 503.

CUSTOMERFive. Five five.

CUSTOMER0139. No.

CUSTOMER013. The gate code is.

CUSTOMERPound 2719. Can anyone come before 2 p.m.? And yes, I agree to the $99 fee.

AGENTThe earliest window offered is 2:00–4:00 PM today; I don’t have an earlier one available. Thank you for agreeing to the $99 fee. I heard your callback number as starting 503-555, but I’m not sure of the last digits. Could you repeat the full number? I have the gate code as #2719.

CUSTOMERSure. My callback number is.

CUSTOMER503.

CUSTOMERFive. Five. Five.

CUSTOMER0193. And the gate code is pound 2719.

CUSTOMERNot 2719. Who will be coming and how will I know when they're on the way? Also, can you give me the ticket number for this visit?

AGENTYou’re booked for today, 2:00–4:00 PM. Your technician is Dana, who will text 503-555-0193 about 30 minutes before arrival. Your ticket number is CPL-58210. The service address is 1480 Larkspur Avenue, unit 4D. I heard the gate code as #2719, but then you said that wasn’t right. Could you confirm the correct code?

CUSTOMERYes. The correct gate code is bound to. 719. That's.

CUSTOMERPound 2719. So Dana will text me 30 minutes before arriving, and the ticket number is.

Insurance

AGENTNorthgate Mutual claims, this is Alex. Are you somewhere safe right now?

CUSTOMERYes, I am safe.

CUSTOMERI am standing next to my car at a cafe near the intersection. I just need some help with the claim.

AGENTI can help with that. Was anyone injured in the accident?

CUSTOMERNo. Nobody is injured.

CUSTOMERIt was just damage to the car.

AGENTI’m glad no one was hurt. Is your car safe to drive? Also, what’s your policy number and the policyholder’s last name?

CUSTOMERNo, the car is not safe to drive.

CUSTOMERThe back bumper is crushed and the rear wheel is rubbing, so I cannot drive it.

CUSTOMERMy policy number is NG4471903.

CUSTOMERNo, sorry, last one is two. The last name is Mehta.

AGENTI’ve verified policy NG4471902. Please don’t drive the car with the wheel rubbing. I can arrange a tow. What’s the café’s address or the nearest intersection, and what phone number can the tow driver call?

CUSTOMERThe nearest intersection. Is Southeast fifth Avenue and Alder Street.

CUSTOMERThe phone number. The tow driver can call is.

CUSTOMER907.

CUSTOMERFour. One six.

CUSTOMERSix. Six. Six. Six.

AGENTI have the intersection as Southeast Fifth Avenue and Alder Street. What city is that in? I heard the phone number as 907-416-66 so far. Could you give me the remaining digits?

CUSTOMERThe city is Portland.

CUSTOMERThe full phone number is 907.

CUSTOMER416.

CUSTOMERSix. Six. Seven six.

AGENTEastside Towing has been dispatched and estimates it will arrive in 45 minutes. The driver can call you at 907-416-6676. The car is scheduled to go to a Northgate partner repair shop at 9th and Morrison. Now, when did the accident happen, and what happened?

CUSTOMERThe accident happened about 20 minutes ago.

CUSTOMERI was stopped at a red light at southeast fifth Avenue and Alder Street. And another car hit the back of my car. The back bumper is crushed and the rear wheel is rubbing. Nobody is hurt. And the police did not come.

AGENTThank you. You were stopped at the light when another car hit you from behind, and police did not attend. What’s the other driver’s name, insurer, and license plate? Also, what time is it there now, so I can record when the accident happened?

CUSTOMERThe other driver's name is Kevin Lund. His insurer is Pacific Shore. The license plate is seven.

CUSTOMERK.

CUSTOMERX like X-ray.

CUSTOMERR.

CUSTOMERTwo one.

CUSTOMERFor the time, here, now is 1:40 p.m..

AGENTI have Kevin Lund, insured by Pacific Shore, with license plate 7KXR21. Is that correct? And what is today’s date?

CUSTOMERAlmost.

CUSTOMERCorrect. The plate is seven.

CUSTOMERK.

CUSTOMERX like X-ray.

CUSTOMERR.

CUSTOMERTwo one.

CUSTOMERFour. Today's date is June 27th.

AGENTThank you for the correction—the plate is 7KXR214. What year is it, so I can record the accident date accurately?

CUSTOMERIt is 2024.

AGENTYour claim is open. The claim number is NGM-904417, and your assigned adjuster is Priya Shah. The claim response says she’ll contact you within one business day.

CUSTOMERAlright, I have noted the claim number NGM904417.

CUSTOMERAnd adjuster Priya Shah kindly let me know. How much do I have to pay for this claim? Is there any deductible? Also, am I eligible for a rental car? And will my premiums go up because of this accident?

AGENTYour policy has a $500 collision deductible. Your rental coverage is up to $40 per day for 30 days, starting from the day the claim is opened. I can’t determine the total amount you’ll have to pay or whether your premium will go up. Priya can discuss those questions with you.

CUSTOMERBut he hit me so I should not have to pay the deductible, right? Also the rental

AGENTI understand why you’d

CUSTOMERcar. Do I just go to the shop at

CUSTOMERnine?

AGENTNot necessarily. Even if the other driver hit you, your $500 collision deductible may apply if you use your own collision coverage. Priya can explain how it applies to your claim. For the rental, your policy provides up to $40 per day for 30 days, starting when the claim was opened. I don’t have information confirming where to pick up a rental or how to arrange one. Please check with Priya before going to the repair shop for a rental.

CUSTOMEROkay. That's clear. I will wait for Priya to contact me about the deductible. And rental car details. Thanks for your help with the claim and towing. Goodbye.

AGENTYou’re welcome.

AGENTTake care, and goodbye.

How these numbers are measured →