Indian-language Saaras and Saarika transcription.
6 integrated models · Official documentation ↗
Sarvam's newest speech recognition model, introduced August 18, 2026 and now its default recommendation. It keeps Saaras v3's five output modes and 22 Indian languages, adds Global English accents, and supports keyterm prompting.
Saaras v4 on Sarvam's asynchronous Batch API, for recordings up to two hours with speaker diarization and up to 50 prompted keyterms. It adds Global English recognition over the Saaras v3 batch model.
Sarvam's Saaras v3 speech recognition for 22 Indian languages plus English, released February 2026 with transcribe, translate, verbatim, transliterate and code-mix output modes. Sarvam reports about 19% WER on IndicVoices; Saaras v4 is now its recommended model.
Asynchronous Batch API for Saaras v3, handling files up to two hours and returning speaker-diarized transcripts with timestamps, which Sarvam offers only through batch. It shares the five output modes of the REST and streaming endpoints.
Sarvam's Bulbul v3 text-to-speech for Indian languages, released February 2026 with 30+ speaker voices in 11 languages (10 Indian plus English) and pace control. Still the API default, though Bulbul v4 Flash is now recommended for new integrations.
Sarvam's earlier Indian-language speech recognition model, released May 2025, which transcribes in the spoken language across 11 languages and handles code-mixed and telephony audio. Now legacy; Saaras v3 replaced it as the API default in June 2026.