ReleasesSep 27, 2026 · 04:561 source

Sarvam AI Releases Saaras V4: A Speech-to-Text Model for All 22 Indian Languages and Global English

From marktechpost.com

AI;DR

Sarvam AI has released Saaras V4, a speech-to-text model covering all 22 Indian languages plus global English. The model pairs an audio encoder with a 3B hybrid state-space decoder and includes keyterm prompting for up to 50 terms, 5 output modes, and streaming with first-token latency under 150 ms. It is available through Sarvam's API at ₹30 per hour.

Read the sourceSign in to suggest

Sources

#llmMarkdown

One ranked edition, every morning

The stories that matter today. Free, cancel anytime.

Telegram @aidr_today