Sarvam AI launches Saaras V4 for 22 Indian languages and English
Sarvam AI has released Saaras V4, a speech-to-text model covering all 22 scheduled Indian languages and English, including global English accents. The company says it achieved state-of-the-art accuracy across the Indian-language set. The model is available through Sarvam’s API with model="saaras:v4"; its weights have not been released, and current SageMaker self-hosting documentation applies only to Saaras v3.
The system combines an audio encoder with a temporal-downsampling adapter and Sarvam-3B, a 3B-parameter hybrid state-space decoder. It supports five output modes: transcribe, verbatim, codemix, translit and translate. V4 also accepts up to 50 keyterms, each up to 64 characters, to bias recognition without guaranteeing their appearance in the transcript.
Sarvam reports a 16.03% WER result on the L5 keyword-prompting setting of IndicContextEval and says Saaras V4’s noisy-audio error rate is less than half that of Deepgram Nova-3 and GPT-4o Transcribe on Kathbath Noisy. These figures are vendor-reported and have not yet been independently reproduced. Streaming is offered over WebSocket with claimed time to first token below 150 milliseconds. Listed pricing is ₹30 per hour, or ₹45 with diarization.
Written by Sakhanda Wire AI from reporting by MarkTechPost. Facts, quotes and full context are in the original article below.