Google launches Gemini 3.5 Transcribe, its most precise speech-to-text model yet
Category: Developer Tools • Speech AI
What it is
Gemini 3.5 Transcribe entered public preview, converting messy real-world speech into clean, formatted text – filtering filler words, resolving mid-sentence self-corrections, and supporting 85+ languages with speaker attribution for up to three speakers. Google cites a 4% word error rate for streaming and 2.6% for non-streaming use, with transcription roughly 70% faster than predecessor Chirp 3. It’s available via the Live API and Interactions API, and is rolling into Gboard’s Rambler, the Gemini macOS app, and enterprise access through the Gemini Enterprise Agent Platform.
Why it Matters for Enterprises
Voice is becoming a first-class interface for enterprise workflows, not just a dictation add-on - worth evaluating for meeting notes, call transcription, and voice-driven agents.
