01
What is Gemini 3.5 Transcribe?
Google introduced Gemini 3.5 Transcribe on August 26, 2026 as a dedicated speech-to-text family built on Gemini's audio understanding. It includes gemini-3.5-transcribe for uploaded recordings and gemini-3.5-transcribe-live for low-latency streaming transcription.
The file-based endpoint targets recordings such as meetings, interviews and call logs, while the Live endpoint streams incremental transcription over WebSockets for voice agents, captioning and interactive dictation.
The models automatically detect more than 85 languages and can switch languages within a session. The recorded-audio model also supports speaker diarization, word-level timestamps and custom vocabulary biasing.
02
Gemini 3.5 Transcribe pricing
Google's current Gemini API pricing page lists a free tier. Paid Gemini 3.5 Transcribe is listed at $2 per million audio input tokens and $12 per million text output tokens, which Google estimates at roughly $0.005 per minute blended.
Gemini 3.5 Transcribe Live is listed at $3.50 per million audio input tokens and $21 per million text output tokens, with Google's estimated blended rate around $0.009 per minute.
03
File transcription vs Live transcription
Use gemini-3.5-transcribe when the audio already exists. It supports up to one hour of audio per request, reduced to 30 minutes when diarization or word-level timestamps are enabled. Speaker diarization supports up to eight speakers, although Google labels attribution for three or more speakers experimental.
Use gemini-3.5-transcribe-live when text needs to arrive while the speaker is talking. It runs through the Gemini Live API over WebSockets and supports sessions up to 10 minutes.
04
Key transcription features
- Automatic language detection across 85+ languages
- Multilingual code-switching
- Smart transcription that can clean up disfluencies and formatting
- Custom vocabulary biasing up to 1,000 terms
- Speaker diarization on recorded audio
- Word-level timestamps on recorded audio
- Real-time WebSocket streaming through Gemini Live API
Google notes that Smart transcription cannot be combined with speaker diarization or word-level timestamps. Those features require verbatim mode.
05
How to use Gemini 3.5 Transcribe
For recorded audio, developers upload an audio file and call the Gemini Interactions API with gemini-3.5-transcribe. Google publishes Python, JavaScript and REST examples in its transcription guide.
For live speech, developers connect through the Gemini Live API with gemini-3.5-transcribe-live and stream continuous audio over WebSockets. Both models are documented in Google AI Studio and the Gemini API.
06
Accuracy and latency
Google says the new model improves on Chirp 3. In its launch announcement, Google reports a 70% improvement in time to final transcription as measured by Artificial Analysis. Google also reports FLEURS word error rates of 5.50% in streaming mode and 5.04% for non-streaming across its selected languages and locales.
These are launch figures reported by Google. They do not establish that Gemini 3.5 Transcribe will beat every competing speech-to-text system across all accents, noise conditions or domains. Production teams should benchmark it on their own recordings.
07
Is Gemini 3.5 Transcribe generally available?
Yes. Google's current Gemini API release notes identify both gemini-3.5-transcribe and gemini-3.5-transcribe-live as generally available (GA) as of August 26, 2026. The model pages, API guides and pricing are live in Google's developer documentation.
08
What can creators and marketers use it for?
Beyond developer voice agents, the model could be useful for meeting notes, multilingual subtitles, podcast transcripts, customer-call analysis, interview transcription and turning spoken ideas into formatted text. Google's Smart transcription mode is particularly relevant when clean readable copy matters more than a verbatim transcript.
For creator workflows, a typical pipeline could be: record video or voice → transcribe with Gemini → turn the transcript into captions, summaries, hooks or repurposed posts with another model.
It is too early to make a definitive ranking against Whisper, Deepgram and other APIs. The useful comparison points will be real-world accuracy, latency, language coverage, diarization, pricing and integration complexity.
09
Bottom line
Gemini 3.5 Transcribe is a meaningful addition to Google's developer stack because it turns transcription into a dedicated Gemini product with both file processing and real-time streaming. Its 85+ languages, code-switching, diarization, timestamps, custom vocabulary and low listed per-minute pricing make it worth testing against real production audio.
Sources
Primary and supporting sources
Facts were rechecked against the linked sources immediately before publication. Pricing, product availability and rollout status can change.