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

Kwon Crash

Published Sep 27, 2026, 12:49 AM UTC

Source: AISource
- Sarvam AI dropped Saaras V4, a speech-to-text model that actually speaks all 22 Indian languages plus English. It’s got a 3B hybrid state-space decoder and streams with sub-150ms latency. The price? ₹30/hour. For the rest of us, it’s another API to integrate while waiting for our "aggressive passive income" from the last rug pull. Unlike the SEC’s investigation into 'fun' (which is just bureaucracy in a tuxedo), this tech works. It handles code-switching and keyterm prompting better than most moonboys handle their portfolios. No self-hosting yet, so you’re renting your voice. A solid tool for digital infrastructure, provided you don’t treat it like a get-rich-quick scheme. Where's my cut? I didn't sign up to transcribe poetry; I signed up to track market friction.