Google announces Gemini 3.5 Transcribe for intelligent speech-to-text

Google DeepMind introduced Gemini 3.5 Transcribe, a speech-to-text model aimed at more intelligent, context-aware transcription. Beyond raw accuracy, the model edits out disfluencies like 'ums' and self-corrections, producing cleaner output for audio-to-text workflows — a practical upgrade for meeting notes, media captioning, and voice-driven applications.

The transcription launch is part of a broader Gemini product wave. Google expanded its Gemini Enterprise platform with tools for lawyers and law firms, naming Freshfields, Cleary, and Weil among early adopters via Google Cloud — a direct enterprise push into the legal vertical, complementing Mistral's and others' domain plays. Gemini Live gained four new capabilities, letting spoken commands hand off multi-step jobs to Spark, which can run for days across Docs, Sheets, and Drive. Google also detailed back-to-school Gemini-in-Workspace features for students and educators.

Competitively, Gemini 3.5 Transcribe pits Google against OpenAI's Whisper lineage and specialized transcription vendors, with the disfluency-editing feature as a differentiator for professional use. Under DMA pressure, Google also agreed to let competing assistants match Gemini on Android — seen by developers as a notable concession on platform openness. The combination of a new transcription model, legal-vertical expansion, and long-running Spark agents shows Google pressing its integrated-productivity advantage. Watch how the legal-vertical rollout performs against incumbents and whether Spark's multi-day agent runs prove reliable in production.