Sarvam AI Releases Saaras V4 Speech-to-Text Model for 22 Indian Languages

Sarvam AI has launched Saaras V4, its latest speech recognition model covering all 22 scheduled Indian languages plus global English. It is available only via API at ₹30 per hour (₹45 with diarization); weights are not public and self-hosting docs still cover v3.

• Decoder is Sarvam-3B, a 3B-parameter hybrid state-space model trained in-house
• Five output modes: transcribe, verbatim, codemix, translit, translate
• WebSocket streaming with time to first token under 150 ms
• Language ID error is 2.9% for top 10 languages and 5.22% for all 22