Saaras V3
Saaras V3 remains available beside V4. It supports 22 Indian languages and Indian English, with the same five output modes. Select its model ID explicitly because an omitted model now selects V4.
Know what timestamps represent
The documented timestamps describe sentence or phrase chunks, not individual words. Speaker-labelled transcripts are available through the batch API.Saaras model guide
Practical use
Practical guidance. Keep V3 as a measured comparison during migration. If an application highlights each spoken word, check whether chunk timestamps can meet that requirement before building the playback interface.
Specifications & access
| Detail | Value and conditions |
|---|---|
| Access | API |
| Model identifiers | saaras:v3 |
| Access | REST, batch and WebSocket streamingSaaras model guide |
Related models
A speech-recognition model for Indian languages and global English with normalized, verbatim, transliterated, code-mixed and translated output.
Launch date not confirmedAPI- Saarika V2.5
Sarvam AI · Saarika
A speech-recognition model that transcribes 11 languages in the language spoken.
Launch date not confirmedDeprecated API; check access detailsAPI