GoogleAudio modelActive
Gemini 3.5 Transcribe
PopularityRank 85 of 875
52out of 100same as last month
Summary
Gemini 3.5 Transcribe is Google’s model for audio files that need structured transcription. The gemini-3.5-transcribe endpoint handles up to one hour of audio per request, recognizes more than 85 languages, and handles code-switching within a recording. Smart Dictation, speaker diarization, word-level timestamps, and custom vocabulary of up to 1,000 terms extend the transcript with structured annotations.
Diarization or word-level timestamps reduce the maximum file length to 30 minutes. Attribution for three or more speakers remains experimental, and custom vocabulary cannot be combined with diarization or timestamps. Gemini 3.5 Transcribe Live is the separate endpoint for WebSocket streaming.
All benchmark results
Every published result with its source, test setup and comparison models.
Show all 37 published benchmark results
Page 1 of 4
Measurements without matching peer values
These measurements have no matching peer values under the same test conditions. Their original values and sources remain available here.
Artificial Analysis AA-WER (non-streaming) (Show measurement, test conditions, and source)
- Source value
- 2.6
- Score
- 2.6
- Metric
- word error rate (WER)
- Unit
- %
- Benchmark version
- AA-WER v2.2
- Category
- audio
- Direction
- Lower is better
- Source type
- independent-evaluator
- Evaluator
- Artificial Analysis
- Status
- active
- Retrieved at
- 2026-10-15
- Methodology
- Independent evaluation. Artificial Analysis directly tested the named API endpoint; audio-duration-weighted WER across AA-AgentTalk (50%), VoxPopuli-Cleaned-AA English (25%), and Earnings22-Cleaned-AA (25%), about eight hours total. Current normalizer AA-WER v2.2.
Google DeepMind FLEURS top-locales non-streaming word error rate (Show measurement, test conditions, and source)
- Source value
- 5.04
- Score
- 5.04
- Metric
- Word error rate across the FLEURS top-locales non-streaming index
- Unit
- %
- Benchmark version
- Google DeepMind FLEURS top-locales non-streaming index, accessed 2026-10-16
- Category
- audio
- Direction
- Lower is better
- Source type
- vendor-reported
- Evaluator
- Google DeepMind
- Status
- active
- Retrieved at
- 2026-10-16
- Methodology
- Vendor-reported result. Google DeepMind reports 5.04% WER for Gemini 3.5 Transcribe in non-streaming mode on its FLEURS top-locales index. The official chart footnote lists ar-EG, bn-IN, cmn-Hans-CN, de-DE, en-US, es-ES, es-US, fr-FR, hi-IN, id-ID, it-IT, ja-JP, ko-KR, mr-IN, nl-NL, pl-PL, pt-BR, ro-RO, ru-RU, ta-IN, te-IN, th-TH, tr-TR, uk-UA, and vi-VN. The source does not specify the aggregation formula or evaluated sample count.
- Context
- Google publishes no evaluated N, confidence interval, per-locale scores, or aggregation formula for this index. This is a vendor-reported aggregate and is not interchangeable with Artificial Analysis or other WER results.
Appen MedTerm-90 overall clinical entity recall (Show measurement, test conditions, and source)
- Source value
- 80.65
- Score
- 80.65
- Metric
- Recall of clinician-validated medical entities
- Unit
- %
- Benchmark version
- MedTerm-90, September 2026 model versions
- Category
- audio
- Direction
- Higher is better
- Sample
- 30017
- Lower confidence bound
- 79.48
- Upper confidence bound
- 81.77
- Source type
- independent-evaluator
- Evaluator
- Appen MedTerm-90
- Status
- active
- Retrieved at
- 2026-10-16
- Methodology
- Independent evaluation. Appen reports the September 2026 Gemini 3.5 Transcribe result on 1413 real physician dictations spanning 90.5 hours and 14 specialties. The reported recall is over 30017 clinician-validated entities; this entity total is the reported count, not the number of recordings. Identical files, entities, thresholds, and judges were used for all models, without vocabulary hints. Appen reports a bootstrap 95% interval.
- Context
- This is clinical-entity recall, not WER or a general transcription score. Results represent September 2026 model versions under close-microphone single-speaker dictation conditions; Appen notes these conditions favor the models and may not generalize to other audio. The report page gives only an overall score, not the advertised per-specialty breakdown.
Superwhisper speech benchmark macro word error rate (Show measurement, test conditions, and source)
- Source value
- 8.2
- Score
- 8.2
- Metric
- macro word error rate (WER) over eight speech datasets
- Unit
- %
- Benchmark version
- Superwhisper speech benchmark, model runs gemini-35-transcribe-full-*-20260827; page captured 2026-10-17
- Category
- speech-to-text / batch transcription
- Direction
- Lower is better
- Source type
- independent-evaluator
- Evaluator
- Superwhisper speech benchmark
- Status
- active
- Retrieved at
- 2026-10-17
- Methodology
- Independent evaluation. Exact product Google Gemini 3.5 Transcribe; hosted batch speech-to-text; evaluator runs the same audio and machines as its other model rows; reported score is macro-average word error rate over 8 named datasets. No medical or realtime route is implied.
- Context
- The page publishes the aggregate and the eight-dataset count, not a clip-level denominator, per-dataset weighting details, confidence interval or immutable serving revision. The run identifiers are preserved as published. This batch/API result is separate from realtime and medical Scribe variants.
Ocular Converse-STT Version 2 pooled WER (Show measurement, test conditions, and source)
- Source value
- 3.13
- Score
- 3.13
- Metric
- pooled word error rate across the eight speaker tracks
- Unit
- %
- Benchmark version
- Converse-STT Version 2; report dated 2026-10-13
- Category
- speech-to-text
- Direction
- Lower is better
- Source type
- independent-evaluator
- Evaluator
- Ocular AI Converse-STT Version 2
- Status
- active
- Retrieved at
- 2026-10-17
- Methodology
- Independent evaluation. Ocular's exact route table identifies Gemini 3.5 Transcribe as the Google OpenRouter route google/gemini-3.5-transcribe with MP3 input and no non-default parameters. The task is batch, single-channel, single-speaker transcription of the full 15-minute tracks, without mixing, diarization or a streaming requirement. The result pools edit counts across four two-person American English conversations, eight speaker tracks and 12,555 reference words, with Whisper's English normalizer whisper-normalizer 0.1.12.
- Context
- This is a small, fixed public sample, not a population estimate. The 12,555-word denominator is not treated as sampleCount. The API-serving revision is not pinned.
Sierra μ-bench overall WER, batch (Show measurement, test conditions, and source)
- Source value
- 7.97
- Score
- 7.97
- Metric
- Unweighted mean of five locale corpus WER values
- Unit
- %
- Benchmark version
- Sierra μ-bench, commit 64d4319be2239441aa470a0036cc2088892c68a3, model score 2026-08-25
- Category
- speech-to-text
- Direction
- Lower is better
- Sample
- 4270
- Source type
- independent-evaluator
- Evaluator
- Sierra μ-bench
- Status
- active
- Retrieved at
- 2026-10-17
- Methodology
- Independent evaluation. Exact pinned model label Gemini-3.5-Transcribe, score date 2026-08-25; 4,270 real customer-service telephone utterances, 8 kHz mono, five locales, human word-level transcripts. Batch API, region us; beam size, language hint, custom vocabulary, noise suppression, domain adaptation, and keyword boosting are provider defaults. Serving revision is not published.
- Context
- The commit-pinned scores.json is the numerical evidence; a mutable dynamic leaderboard is not substituted for this snapshot. significantWer means significant word error rate, not UER. No confidence interval or immutable provider serving revision is published.
Sierra μ-bench significant word error rate, batch (Show measurement, test conditions, and source)
- Source value
- 14.93
- Score
- 14.93
- Metric
- Sierra significantWer, semantically significant word error rate
- Unit
- %
- Benchmark version
- Sierra μ-bench, commit 64d4319be2239441aa470a0036cc2088892c68a3, model score 2026-08-25
- Category
- speech-to-text
- Direction
- Lower is better
- Sample
- 4270
- Source type
- independent-evaluator
- Evaluator
- Sierra μ-bench
- Status
- active
- Retrieved at
- 2026-10-17
- Methodology
- Independent evaluation. Exact pinned model label Gemini-3.5-Transcribe, score date 2026-08-25; 4,270 real customer-service telephone utterances, 8 kHz mono, five locales, human word-level transcripts. Batch API, region us; beam size, language hint, custom vocabulary, noise suppression, domain adaptation, and keyword boosting are provider defaults. Serving revision is not published.
- Context
- The commit-pinned scores.json is the numerical evidence; a mutable dynamic leaderboard is not substituted for this snapshot. significantWer means significant word error rate, not UER. No confidence interval or immutable provider serving revision is published.
Superwhisper Gemini 3.5 Transcribe dataset WER: AMI Meetings (Show measurement, test conditions, and source)
- Source value
- 10.7
- Score
- 10.7
- Metric
- Dataset-specific word error rate (WER)
- Unit
- %
- Benchmark version
- Superwhisper speech benchmark, updated 2026-09-03; row run token gemini-35-transcribe-full-tail-20260827
- Direction
- Lower is better
- Source type
- independent-evaluator
- Evaluator
- Superwhisper
- Status
- active
- Retrieved at
- 2026-10-18
- Methodology
- Independent evaluation. Google Gemini 3.5 Transcribe, cloud/public API, dataset AMI Meetings; exact source row run token gemini-35-transcribe-full-tail-20260827. Superwhisper says cloud results use the same audio and machines; its page does not pin an API revision or request settings.
- Context
- These are dataset-specific WER slices from one Superwhisper evaluation, not eight independent model experiments. The source does not publish dataset-specific sample counts, confidence intervals, an immutable API revision, or request settings. Superwhisper is the evaluator and also publishes its own competing products; this is evaluator-reported, not a Google-reported score.
Superwhisper Gemini 3.5 Transcribe dataset WER: Common Voice, Australian English (Show measurement, test conditions, and source)
- Source value
- 4.6
- Score
- 4.6
- Metric
- Dataset-specific word error rate (WER)
- Unit
- %
- Benchmark version
- Superwhisper speech benchmark, updated 2026-09-03; row run token gemini-35-transcribe-full-canonical-20260827
- Direction
- Lower is better
- Source type
- independent-evaluator
- Evaluator
- Superwhisper
- Status
- active
- Retrieved at
- 2026-10-18
- Methodology
- Independent evaluation. Google Gemini 3.5 Transcribe, cloud/public API, dataset Common Voice, Australian English; exact source row run token gemini-35-transcribe-full-canonical-20260827. Superwhisper says cloud results use the same audio and machines; its page does not pin an API revision or request settings.
- Context
- These are dataset-specific WER slices from one Superwhisper evaluation, not eight independent model experiments. The source does not publish dataset-specific sample counts, confidence intervals, an immutable API revision, or request settings. Superwhisper is the evaluator and also publishes its own competing products; this is evaluator-reported, not a Google-reported score.
Superwhisper Gemini 3.5 Transcribe dataset WER: Common Voice, spontaneous English (Show measurement, test conditions, and source)
- Source value
- 9.1
- Score
- 9.1
- Metric
- Dataset-specific word error rate (WER)
- Unit
- %
- Benchmark version
- Superwhisper speech benchmark, updated 2026-09-03; row run token gemini-35-transcribe-full-canonical-20260827
- Direction
- Lower is better
- Source type
- independent-evaluator
- Evaluator
- Superwhisper
- Status
- active
- Retrieved at
- 2026-10-18
- Methodology
- Independent evaluation. Google Gemini 3.5 Transcribe, cloud/public API, dataset Common Voice, spontaneous English; exact source row run token gemini-35-transcribe-full-canonical-20260827. Superwhisper says cloud results use the same audio and machines; its page does not pin an API revision or request settings.
- Context
- These are dataset-specific WER slices from one Superwhisper evaluation, not eight independent model experiments. The source does not publish dataset-specific sample counts, confidence intervals, an immutable API revision, or request settings. Superwhisper is the evaluator and also publishes its own competing products; this is evaluator-reported, not a Google-reported score.
Superwhisper Gemini 3.5 Transcribe dataset WER: Earnings 22 (Show measurement, test conditions, and source)
- Source value
- 8.8
- Score
- 8.8
- Metric
- Dataset-specific word error rate (WER)
- Unit
- %
- Benchmark version
- Superwhisper speech benchmark, updated 2026-09-03; row run token gemini-35-transcribe-full-tail-20260827
- Direction
- Lower is better
- Source type
- independent-evaluator
- Evaluator
- Superwhisper
- Status
- active
- Retrieved at
- 2026-10-18
- Methodology
- Independent evaluation. Google Gemini 3.5 Transcribe, cloud/public API, dataset Earnings 22; exact source row run token gemini-35-transcribe-full-tail-20260827. Superwhisper says cloud results use the same audio and machines; its page does not pin an API revision or request settings.
- Context
- These are dataset-specific WER slices from one Superwhisper evaluation, not eight independent model experiments. The source does not publish dataset-specific sample counts, confidence intervals, an immutable API revision, or request settings. Superwhisper is the evaluator and also publishes its own competing products; this is evaluator-reported, not a Google-reported score.
Superwhisper Gemini 3.5 Transcribe dataset WER: Earnings 22 (with vocabulary) (Show measurement, test conditions, and source)
- Source value
- 15.8
- Score
- 15.8
- Metric
- Dataset-specific word error rate (WER)
- Unit
- %
- Benchmark version
- Superwhisper speech benchmark, updated 2026-09-03; row run token gemini-35-transcribe-full-canonical-20260827
- Direction
- Lower is better
- Source type
- independent-evaluator
- Evaluator
- Superwhisper
- Status
- active
- Retrieved at
- 2026-10-18
- Methodology
- Independent evaluation. Google Gemini 3.5 Transcribe, cloud/public API, dataset Earnings 22 (with vocabulary); exact source row run token gemini-35-transcribe-full-canonical-20260827. Superwhisper says cloud results use the same audio and machines; its page does not pin an API revision or request settings.
- Context
- These are dataset-specific WER slices from one Superwhisper evaluation, not eight independent model experiments. The source does not publish dataset-specific sample counts, confidence intervals, an immutable API revision, or request settings. Superwhisper is the evaluator and also publishes its own competing products; this is evaluator-reported, not a Google-reported score.
Page 1 of 4
Popularity compared
How often the model is searched for, used and mentioned, compared with every model of its category.
Popularity index
0 to 100, higher is more popular
20 of 83 models
pyannote Community-1, pyannote: 73
Nemotron-3-Diarization, NVIDIA: 57
Gemini 3.5 Transcribe, Google: 52
Confucius4-R2T2, NetEase Youdao: 52
Granite Speech 5.0 470M TurboCTC, IBM: 52
Kokoro-82M, hexgrad: 51
Nova-3, Deepgram: 51
Fish Audio S2 Pro, Fish Audio: 51
NVIDIA Nemotron 3.5 Transcribe, NVIDIA: 50
Granite Speech 5.0 470M TurboCTC NC, IBM: 50
Orukeet, Oruk: 49
Lyria 3.5, Google: 47
Gemini 3.5 Live Translate, Google: 46
Phonon-2, Fermion Research: 44
Aura-2, Deepgram: 44
Gemini 3.1 Flash TTS, Google: 44
MiniMax-Music3, MiniMax: 41
Audio8-ASR-Infinite, Edge0 AI: 41
GPT-Realtime 2.1, OpenAI: 41
Eleven v3, ElevenLabs: 39
Gemini 3.5 Transcribe
As of November 2026 · Source: gradually.ai popularity index
Over time
Weekly values from October 18, 2026 to November 5, 2026
Gemini 3.5 TranscribeConfucius4-R2T2Granite Speech 5.0 470M TurboCTCKokoro-82MNova-3
All specifications
Architecture
Input and output
Watermarking and provenance
AI watermarking refers to invisible signals embedded in the output. Visible marks and provenance metadata are listed separately. Unknown means there is insufficient evidence for that particular field.
- AI watermarking
- Unknown
- no source
- Watermarking technique
- Not known
- no source
Licence and access
API
Further details
- Audio duration
- Up to 1 hour; up to 30 minutes with diarization or word timestamps
- Source: Audio duration
Sources and data date
Show all 24 sources
Benchmark sources at a glance
Which independent evaluators tested this model. This is coverage, not a quality ranking.
| Evaluator | Test series | Composite indices | Recorded active measurements |
|---|---|---|---|
| Artificial Analysis | 0 | 0 | 1 |
Without vendor claims and historical or disputed results; those stay labelled in the single values. 1 measurements are not yet assigned to a test series.