BENCHMARKSTT24 MODELS RANKEDLAST 30 DAYS

Time to Final Segment (TTFS)

Name: Time to Final Segment voice AI benchmark results
Creator: Coval

Time to Final Segment (TTFS) is how long a speech-to-text model takes to deliver its final transcript after the speaker stops talking.

Current STT leader#1 / 24: 64ms; STT RT v5via Soniox

How it is calculated

TTFS = final-transcript timestamp − shared end-of-speech timestamp.

How to read it

Unit: ms
Better: Lower
Period: Rolling 30 days
Cadence: Re-measured daily

Speech-to-Text models on Time to Final Segment

Full STT dashboard

Time to Final Segment — every measured STT modelMilliseconds · lower is better · 30-day average

#1STT RT v564 ms
#2Parakeet TDT 0.6B v370 ms
#3STT 183 ms
#4Nova 399 ms
#5Nova 2101 ms
#6Ink 2108 ms
#7Scribe v2 Realtime120 ms
#8Universal 3.5 Pro146 ms
#9Defaultvia Azure155 ms
#10Whisper Large v3180 ms
#11Velma 2 STT Streaming191 ms
#12Grok STT199 ms

Show all 24 models

#13Pulse205 ms
#14Defaultvia Speechmatics223 ms
#15Defaultvia Gradium249 ms
#16resonant-1288 ms
#17Enhanced341 ms
#18Voxtral Mini Transcribe Realtime 2602356 ms
#19GPT Realtime Whisper559 ms
#20GPT-4o mini Transcribe623 ms
#21Solaria 1680 ms
#22GPT-4o Transcribe742 ms
#23Chirp 2811 ms
#24Chirp 3813 ms

median of all models · 202 ms

STT models over the last 30 days, ranked on Time to Final Segment.
#	Model	Host	TTFS	TTFT	Samples
1	STT RT v5	Soniox	64 ms	1533 ms	29,322
2	Parakeet TDT 0.6B v3	Together AI	70 ms	1210 ms	29,161
3	STT 1	Inworld AI	83 ms	1468 ms	29,280
4	Nova 3	Deepgram	99 ms	1434 ms	29,290
5	Nova 2	Deepgram	101 ms	1436 ms	29,145
6	Ink 2	Cartesia	108 ms	1812 ms	29,288
7	Scribe v2 Realtime	ElevenLabs	120 ms	2157 ms	29,313
8	Universal 3.5 Pro	AssemblyAI	146 ms	1030 ms	29,280
9	Default	Azure	155 ms	1792 ms	6,682
10	Whisper Large v3	Together AI	180 ms	1241 ms	29,171
11	Velma 2 STT Streaming	Modulate	191 ms	1576 ms	17,895
12	Grok STT	xAI	199 ms	—	29,322
13	Pulse	Smallest	205 ms	2072 ms	29,313
14	Default	Speechmatics	223 ms	1473 ms	29,323
15	Default	Gradium	249 ms	1979 ms	28,235
16	resonant-1	Reson8	288 ms	—	17,066
17	Enhanced	Speechmatics	341 ms	1536 ms	29,322
18	Voxtral Mini Transcribe Realtime 2602	Mistral	356 ms	1828 ms	29,175
19	GPT Realtime Whisper	OpenAI	559 ms	1823 ms	29,286
20	GPT-4o mini Transcribe	OpenAI	623 ms	—	29,318
21	Solaria 1	Gladia	680 ms	1703 ms	26,356
22	GPT-4o Transcribe	OpenAI	742 ms	—	29,318
23	Chirp 2	Google	811 ms	6000 ms	29,237
24	Chirp 3	Google	813 ms	5999 ms	29,319
—	Flux	Deepgram	—	1089 ms	29,354
—	Flux Multilingual	Deepgram	—	1175 ms	29,344
—	Nemotron 3.5 ASR Streaming	Together AI	—	1540 ms	29,193
—	Universal Streaming	AssemblyAI	—	1510 ms	29,297

Under-sampled models are excluded; tied models share a place. Dotted WER values split into substitutions, deletions and insertions on hover or tap.

Tail latency

The distribution behind each average, ranked by median TTFS.

Time to Final Segment distribution per modelMilliseconds · shared axis · last 30 days

STT RT v5p50 62 ms · p99 115 ms
Parakeet TDT 0.6B v3p50 62 ms · p99 167 ms
STT 1p50 72 ms · p99 148 ms
Velma 2 STT Streamingp50 86 ms · p99 1644 ms
Nova 3p50 96 ms · p99 194 ms
Nova 2p50 96 ms · p99 217 ms
Ink 2p50 106 ms · p99 169 ms
Scribe v2 Realtimep50 111 ms · p99 297 ms
Whisper Large v3p50 112 ms · p99 1641 ms
Universal 3.5 Prop50 118 ms · p99 458 ms
Defaultvia Azurep50 158 ms · p99 265 ms
Pulsep50 193 ms · p99 373 ms
Grok STTp50 193 ms · p99 336 ms
Defaultvia Speechmaticsp50 215 ms · p99 319 ms
Defaultvia Gradiump50 248 ms · p99 368 ms
resonant-1p50 273 ms · p99 473 ms
Voxtral Mini Transcribe Realtime 2602p50 310 ms · p99 1025 ms
Enhancedp50 321 ms · p99 648 ms
Solaria 1p50 393 ms · p99 6796 ms
GPT Realtime Whisperp50 560 ms · p99 952 ms
GPT-4o mini Transcribep50 578 ms · p99 1371 ms
GPT-4o Transcribep50 714 ms · p99 1441 ms
Chirp 2p50 767 ms · p99 1586 ms
Chirp 3p50 789 ms · p99 1603 ms

band p25–p75 · tick p50 · whisker to p99 with p90 and p95 stops

Last 30 days

Daily median for the current top 5 · gaps are days without qualifying runs.

Time to Final Segment — current leadersDaily p50 per model · UTC days

Nova 2Nova 3Parakeet TDT 0.6B v3STT 1STT RT v5

Time to Final Segment by dataset

The overall average split by test condition — where each model holds up and where it degrades.

Time to Final Segment for every STT model, per dataset, over the last 30 days.
#	Model	Host	All datasets	WildASR clean	PipeCat (production)	WildASR accents	WildASR clipping	WildASR far-field	WildASR noise gaps	WildASR phone codec	WildASR reverb
1	STT RT v5	Soniox	64 ms	66 ms	64 ms	66 ms	62 ms	61 ms	62 ms	61 ms	63 ms
2	Parakeet TDT 0.6B v3	Together AI	70 ms	66 ms	70 ms	73 ms	71 ms	70 ms	67 ms	72 ms	71 ms
3	STT 1	Inworld AI	83 ms	87 ms	81 ms	79 ms	87 ms	91 ms	88 ms	67 ms	81 ms
4	Nova 3	Deepgram	99 ms	103 ms	99 ms	96 ms	96 ms	96 ms	94 ms	94 ms	94 ms
5	Nova 2	Deepgram	101 ms	99 ms	106 ms	93 ms	95 ms	98 ms	95 ms	93 ms	93 ms
6	Ink 2	Cartesia	108 ms	108 ms	108 ms	110 ms	107 ms	110 ms	107 ms	111 ms	110 ms
7	Scribe v2 Realtime	ElevenLabs	120 ms	121 ms	119 ms	119 ms	127 ms	123 ms	122 ms	120 ms	122 ms
8	Universal 3.5 Pro	AssemblyAI	146 ms	140 ms	145 ms	173 ms	147 ms	151 ms	145 ms	143 ms	147 ms
9	Default	Azure	155 ms	150 ms	163 ms	116 ms	153 ms	152 ms	159 ms	154 ms	147 ms
10	Whisper Large v3	Together AI	180 ms	178 ms	186 ms	177 ms	166 ms	166 ms	156 ms	185 ms	176 ms
11	Velma 2 STT Streaming	Modulate	191 ms	170 ms	200 ms	180 ms	184 ms	182 ms	194 ms	191 ms	205 ms
12	Grok STT	xAI	199 ms	197 ms	202 ms	164 ms	251 ms	200 ms	200 ms	186 ms	177 ms
13	Pulse	Smallest	205 ms	205 ms	202 ms	209 ms	211 ms	215 ms	202 ms	215 ms	216 ms
14	Default	Speechmatics	223 ms	221 ms	220 ms	244 ms	234 ms	228 ms	223 ms	223 ms	231 ms
15	Default	Gradium	249 ms	250 ms	241 ms	318 ms	241 ms	247 ms	243 ms	274 ms	253 ms
16	resonant-1	Reson8	288 ms	286 ms	293 ms	262 ms	283 ms	286 ms	288 ms	285 ms	283 ms
17	Enhanced	Speechmatics	341 ms	341 ms	339 ms	347 ms	349 ms	342 ms	343 ms	340 ms	342 ms
18	Voxtral Mini Transcribe Realtime 2602	Mistral	356 ms	353 ms	353 ms	346 ms	362 ms	359 ms	354 ms	382 ms	369 ms
19	GPT Realtime Whisper	OpenAI	559 ms	553 ms	560 ms	567 ms	566 ms	559 ms	551 ms	559 ms	567 ms
20	GPT-4o mini Transcribe	OpenAI	623 ms	622 ms	628 ms	557 ms	628 ms	633 ms	630 ms	613 ms	618 ms
21	Solaria 1	Gladia	680 ms	704 ms	647 ms	735 ms	733 ms	726 ms	719 ms	748 ms	670 ms
22	GPT-4o Transcribe	OpenAI	742 ms	744 ms	751 ms	638 ms	743 ms	741 ms	750 ms	743 ms	737 ms
23	Chirp 2	Google	811 ms	807 ms	826 ms	704 ms	803 ms	802 ms	823 ms	818 ms	783 ms
24	Chirp 3	Google	813 ms	817 ms	829 ms	699 ms	778 ms	790 ms	798 ms	833 ms	794 ms

Ranked on the all-datasets average; a dash means the model was not measured on that dataset in the last 30 days. Cell shading deepens toward each column’s highest value.

About TTFS

Why it matters

Applications often wait for a final transcript before acting on a completed turn. A model may stream partial text quickly but take longer to mark the segment final.

Coval uses TTFS as the primary speech-to-text ranking metric because it measures when the final transcript becomes available after speech ends.

How Coval measures it

One shared voice-activity model identifies the end of speech, and every provider's timer starts from that point. Provider-specific endpointing decisions are therefore excluded from the comparison.

Every model receives the same audio, and connection setup is excluded consistently across providers.

Caveats and interpretation

A production agent may add its own endpointing delay before the benchmark's shared end-of-speech instant.
Providers expose different finalization semantics; Coval uses the event treated as final by the published integration for each endpoint.

The datasets behind it

Every model runs the same fixed inputs, so a gap in TTFS is the model's doing — not the test's.

Same datasets, prompts and metric definitions for every model, measured by Coval’s open-source runner. Full methodology on the overview.

Evaluate your own voice agent

Use Coval to test your production configuration, prompts and calls—not only the public benchmark endpoints.

Book a Demo