GoogleElevenLabsAudio matchupSpeech to text
Gemini 3.5 Transcribe Livevs.Scribe v2 Realtime
Which model is better?
Gemini 3.5 Transcribe Live and Scribe v2 Realtime each lead in different tests.
Pick Gemini 3.5 Transcribe Live for
Pick Scribe v2 Realtime for
Shared tests
Independent evaluators and leaderboards only, the newest shared snapshot per evaluator. With 95 % ranges a lead counts when they do not overlap; for hit rates, when it holds in a 95 % test. Without any uncertainty figure at most “slightly ahead”.
Coval
depends on the subtestScribe v2 Realtime is clearly ahead in 0 of 18 subtests and slightly in 8, Gemini 3.5 Transcribe Live in 0 clearly and slightly in 6. 4 are level.
Coval Time to Final Segment (TTFS) mean
lower is better, as of November 6, 2026
Coval Time to Final Segment (TTFS) p50
lower is better, as of November 6, 2026
Coval Time to Final Segment (TTFS) p95
lower is better, as of November 6, 2026
Show all 18 subtests
| Subtest | Gemini 3.5 Transcribe Live | Scribe v2 Realtime | Verdict |
|---|---|---|---|
| Coval Time to Final Segment (TTFS) mean (lower is better) | 380 ms | 119 ms | Scribe v2 Realtime slightly ahead |
| Coval Time to Final Segment (TTFS) p50 (lower is better) | 325 ms | 111 ms | Scribe v2 Realtime slightly ahead |
| Coval Time to Final Segment (TTFS) p95 (lower is better) | 493 ms | 179 ms | Scribe v2 Realtime slightly ahead |
| Coval Time to First Token (TTFT) maximum (lower is better) | 12,993 ms | 4,960 ms | Scribe v2 Realtime slightly ahead |
| Coval RTF maximum (lower is better) | 3.12 | 1.96 | Scribe v2 Realtime slightly ahead |
| Coval pooled corpus Word Error Rate (WER) (lower is better) | 2.9 % | 4.2 % | Gemini 3.5 Transcribe Live slightly ahead |
| Coval Time to First Token (TTFT) p50 (lower is better) | 1,471 ms | 2,111 ms | Gemini 3.5 Transcribe Live slightly ahead |
| Coval Time to First Token (TTFT) mean (lower is better) | 1,528 ms | 2,149 ms | Gemini 3.5 Transcribe Live slightly ahead |
| Coval Word Error Rate (WER) per-sample p95 (lower is better) | 16.7 % | 21.4 % | Gemini 3.5 Transcribe Live slightly ahead |
| Coval Time to First Token (TTFT) p95 (lower is better) | 2,416 ms | 2,916 ms | Gemini 3.5 Transcribe Live slightly ahead |
| Coval Time to Final Segment (TTFS) maximum (lower is better) | 5,018 ms | 5,714 ms | Gemini 3.5 Transcribe Live slightly ahead |
| Coval AudioToFinal maximum (lower is better) | 18.79 s | 16.82 s | Scribe v2 Realtime slightly ahead |
| Coval RTF mean (lower is better) | 1.06 | 1.02 | Scribe v2 Realtime slightly ahead |
| Coval AudioToFinal mean (lower is better) | 8.31 s | 8.06 s | Scribe v2 Realtime slightly ahead |
| Coval RTF p50 (lower is better) | 1.04 | 1.01 | level |
| Coval AudioToFinal p50 (lower is better) | 8.67 s | 8.43 s | level |
| Coval Word Error Rate (WER) per-sample maximum (lower is better) | 100.0 % | 100.0 % | level |
| Coval Word Error Rate (WER) per-sample p50 (lower is better) | 0.0 % | 0.0 % | level |
Source · latest November 6, 2026
Pipecat STT Benchmark
Gemini 3.5 Transcribe Live slightly aheadGemini 3.5 Transcribe Live is clearly ahead in 0 of 2 subtests and slightly in 2, Scribe v2 Realtime in none.
Pipecat STT Benchmark semantic pooled WER
lower is better, as of October 16, 2026
Pipecat STT Benchmark semantic WER mean
lower is better, as of October 16, 2026
Source · latest October 16, 2026
Artificial Analysis
Scribe v2 Realtime slightly aheadArtificial Analysis AA-WER Streaming (final transcription)
lower is better, as of October 15, 2026
Source · latest October 15, 2026
Vendor figures
Figures from vendor reports that cover both models. They do not count toward the verdict.
Show 6 vendor figures
| Figure | Source | Gemini 3.5 Transcribe Live | Scribe v2 Realtime |
|---|---|---|---|
| Accented English WER, Mean 4 sets (lower is better) | AssemblyAI speech benchmark | 9.4 % | 10.1 % |
| AssemblyAI normalized medical entity error rate mean across three sets (lower is better) | AssemblyAI speech benchmark | 8.6 % | 7.0 % |
| AssemblyAI normalized WER mean across three noise conditions (lower is better) | AssemblyAI speech benchmark | 8.5 % | 7.8 % |
| Code-switching WER, Mean 5 pairs (lower is better) | AssemblyAI speech benchmark | 10.2 % | 13.0 % |
| Long-form English WER, Mean 3 sets (lower is better) | AssemblyAI speech benchmark | 9.4 % | 17.1 % |
| Short-form English WER, Mean 3 sets (lower is better) | AssemblyAI speech benchmark | 5.0 % | 5.1 % |
All data side by side
Prices are verbatim from the provider, because audio models bill per minute, character or token.
| Gemini 3.5 Transcribe Live | Scribe v2 Realtime | |
|---|---|---|
| Task | Speech to text | Speech to text |
| Developer | ElevenLabs | |
| Status | Active | Active |
| Weights | proprietary | proprietary |
| Runs locally | no | no |
| Price | $3.50 per 1M audio tokens ($0.005 per audio minute) | n/a |
| Popularity (out of 100) | 35 | 29 |