What it is
Gemini 3.5 Transcribe and Gemini 3.5 Transcribe Live became generally available on 26 August 2026. They are dedicated speech-to-text models built on Gemini's audio understanding, separate from the general Gemini chat models, and billed by audio and text tokens rather than by the minute.
The file model takes up to an hour of audio per request, or 30 minutes when diarization or word timestamps are switched on. It detects 85+ languages including mid-sentence switching, labels up to eight speakers, gives word-level timestamps, cleans up filler words in a "Smart" mode and accepts up to 1,000 custom vocabulary terms. The Live model streams over WebSockets with interim and final results, in sessions of up to 10 minutes, but without diarization or timestamps.
Our review
Gemini 3.5 Transcribe is a good model in a market that has moved quickly past it. Artificial Analysis measures it at 2.6% AA-WER, ninth of 62 models on its non-streaming board, at $5 per 1,000 minutes and about 89 times real-time. Microsoft's MAI-Transcribe-2 is more accurate at 2.0% for $1.67, ElevenLabs Scribe v2 manages 2.2% for $3.67, while OpenAI's GPT Transcribe (3.3%) and Deepgram Nova-3 (5.2%) trail it.
The feature set is broad, but the options collide. Google's own model page says word timestamps degrade accuracy, custom vocabulary is rejected alongside either diarization or timestamps, and attribution for three or more speakers is experimental. A meeting-notes product that wants named speakers, timestamps and the client's product names has to choose two of the three, or send the audio twice.
Price is simple and steady. At Google's estimate of 25 audio tokens per second and 175 text tokens per minute, the file model costs about half a cent a minute and Live about 0.9 cents, but neither supports the Batch or Flex discounts that Gemini chat models get, so volume does not get cheaper. The free tier is generous, and Google's terms say content sent on it is used to improve Google products; paid-tier content is not.
The strongest argument is consolidation. A team already billing Gemini through Google Cloud gets file and streaming transcription under one key, one data agreement and one invoice, with code-switching across 85+ languages detected per utterance. For a pure transcription workload chosen on accuracy per dollar, MAI-Transcribe-2 now sets the bar.
Where it earns its place
A 200-seat contact centre already on Google Cloud
Transcribe every call for QA and compliance search
Send recordings to the file model with diarization on, keeping each under 30 minutes. At 500,000 minutes a month that is about $2,500 on the paid tier; MAI-Transcribe-2 at Artificial Analysis's $1.67 per 1,000 minutes would be about $835.
A mobile engineer building voice dictation into a notes app
Show live captions as the user speaks
Stream 100 ms audio chunks to Transcribe Live over WebSockets and render interim results, swapping in finals. Sessions end at 10 minutes, so long dictation needs reconnect logic; cost is about 0.9 cents a minute.
A legal-tech product manager in a multilingual market
Transcribe client interviews that switch between English and Hindi
Utterance-level language detection handles the switches without a language hint. Firm and product names need custom vocabulary, which means turning off diarization and timestamps for that pass.
Before you commit
- Setup
- Get a Gemini API key in AI Studio, upload audio through the Files API and call gemini-3.5-transcribe; the docs give Python, JavaScript, Go and REST samples.
- Integrations
- Gemini API and Google GenAI SDKs; Live transcription over the Live API WebSocket with configurable voice-activity detection. No caching, Batch, Flex or function calling.
- Data and privacy
- On the free tier Google may use content to improve its products; on the paid tier it does not. Apps serving users in the EEA, Switzerland or the UK must use paid services.
- Cost at scale
- About $5 per 1,000 minutes for files and $9 for Live, with no volume or batch discount. Artificial Analysis lists Scribe v2 at $3.67 and MAI-Transcribe-2 at $1.67 per 1,000 minutes.
- Limits
- One hour per file request (30 minutes with diarization or timestamps), 10-minute Live sessions, up to eight speakers with three or more experimental, and custom vocabulary incompatible with diarization and timestamps.
What's good
- 2.6% AA-WER at about $5 per 1,000 minutes
- File and WebSocket streaming models under one API
- 85+ languages with mid-sentence code-switching
- Up to 1,000 custom vocabulary terms
What's not
- Custom vocabulary cannot be combined with diarization or timestamps
- MAI-Transcribe-2 and Scribe v2 are more accurate and cheaper
- No Batch or Flex discount
- Live sessions capped at 10 minutes, no diarization
Best for
Teams already on the Gemini API who want file and streaming transcription from one vendor at about half a cent a minute.
Alternatives
Gemini 3.5 Transcribe: common questions
- Is Gemini 3.5 Transcribe worth it?
- Accurate and cheap enough for most transcription, yet no longer the obvious buy on either measure. We rate it 4.2 out of 5.
- How much does Gemini 3.5 Transcribe cost?
- Transcribe about $0.005 per minute ($2 input / $12 output per 1M tokens); Transcribe Live about $0.009 per minute ($3.50 / $21). Free tier in AI Studio and the API.
- What are the downsides of Gemini 3.5 Transcribe?
- Custom vocabulary cannot be combined with diarization or timestamps; mAI-Transcribe-2 and Scribe v2 are more accurate and cheaper; no Batch or Flex discount; live sessions capped at 10 minutes, no diarization.
- Who is Gemini 3.5 Transcribe best for?
- Teams already on the Gemini API who want file and streaming transcription from one vendor at about half a cent a minute.
- How much does Gemini 3.5 Transcribe cost for a team or in production?
- About $5 per 1,000 minutes for files and $9 for Live, with no volume or batch discount. Artificial Analysis lists Scribe v2 at $3.67 and MAI-Transcribe-2 at $1.67 per 1,000 minutes.
- Is Gemini 3.5 Transcribe safe for company data?
- On the free tier Google may use content to improve its products; on the paid tier it does not. Apps serving users in the EEA, Switzerland or the UK must use paid services.
- What are the best Gemini 3.5 Transcribe alternatives?
- The closest alternatives we have reviewed are ElevenLabs Scribe v2, GPT Transcribe and Deepgram Nova-3.
Answers checked Oct 22, 2026.
Make Gemini 3.5 Transcribe? Show this rating on your site
