Speech-to-text AI models: two built purely for transcription

Speech-to-text

Services found: 2

#	AI model	Capabilities	API $/1M (in / out)	Context window	Price / mo	Output $/1M	Overview	Visit site
1	Gemini 3.5 Transcribe Google				API only		View
2	Gemini 3.5 Transcribe Live Google				API only		View

Browse by need

Reasoning models169 models

Coding assistants17 models

Vision and images156 models

Long context 1M+82 models

Free tier available80 models

Open weights88 models

Two purpose-built transcription models are listed: Gemini 3.5 Transcribe for recorded files and Gemini 3.5 Transcribe Live for streaming. Both handle automatic language detection, speaker diarisation, word-level timestamps and custom vocabulary.

Transcription is priced by audio duration or audio tokens rather than by words, which makes it far cheaper than reasoning over the same recording. The live variant costs more per minute and exists where text must appear as the person speaks: captions, live notes, voice commands. For meeting records, batch transcription afterwards is the economical path.

Priced by durationMinutes and audio tokens, not words — far cheaper than reasoning over the same file.

Diarisation includedSpeaker separation and word timestamps come as standard here.

Live costs moreStreaming carries a premium over batch transcription of the same recording.

Models with this capability

The 2 models below all support this capability. Prices are input and output per 1M tokens from the vendor official price list; for dirhams multiply by 3.6725. Tap a card to open the model page.

Gemini 3.5 Transcribe

Google DeepMind

price on request

Gemini 3.5 Transcribe Live

Google DeepMind

price on request

How to choose

Choose batch unless the text must appear as the person speaks.
Test on your real audio: accents, overlap and background noise decide accuracy.
Supply a custom vocabulary for names, products and jargon.
Check language detection if your recordings are mixed.

Common mistakes

Using a full audio model for plain transcriptsYou pay reasoning rates for a job a transcription model does better.

Skipping custom vocabularyProper nouns are where accuracy visibly fails.

Who this suits

MeetingsRecords and searchable archives without note-taking.

MediaSubtitles and transcripts at volume.

ComplianceCall records that must be retained and searchable.

Related filters

Audio understanding25 models that interpret speech and sound

Realtime voice7 models for live voice conversation

TranslationFaithful translation with terminology control

MultilingualAll 239 models work beyond English

All 239 modelsThe full rating with filters and prices

Frequently asked questions

File or streaming?

Gemini 3.5 Transcribe handles recorded files; Transcribe Live streams as the person speaks.

How is transcription priced?

By audio duration or audio tokens rather than words — far cheaper than reasoning over the same recording.

Which is right for meetings?

Batch transcription afterwards. Live costs more per minute and exists for captions and voice commands.

How accurate is it on accented English?

Good but not uniform. Test with your own speakers before committing.

Can it handle two languages in one recording?

Automatic detection helps, though switching mid-sentence remains hard.

Ratings 0

Avg. score 0

Rate this page

Help us improve