Google introduces Gemini 3.5 Transcribe, a high-precision speech-to-text model with 70% latency improvement, enabling voice editing and multilingual support across Gboard and Chrome.
Google launched Gemini 3.5 Transcribe on August 26, 2026, positioning it as its most accurate speech-to-text model yet. The AI system converts raw audio into polished, formatted text while handling self-corrections, filler word removal, and natural voice editing. Early integrations include Gboard Rambler on Android and the Gemini app for macOS, with Chrome support planned.
Unlike traditional models, Gemini 3.5 Transcribe captures natural speech patterns to interpret intent and recognize custom vocabulary. It can process self-corrections like 'let’s meet Tuesday—no, Wednesday' and auto-format text for clarity. The model also supports function calling, enabling delegation of tasks like image generation to other Gemini systems.
Performance metrics show a 70% reduction in transcription latency compared to Chirp 3, Google’s 2025 baseline. On the FLEURS benchmark, it achieves a 5.50% word error rate (WER) in streaming mode and 5.04% in non-streaming scenarios across 85+ languages. Artificial Analysis confirmed these improvements, highlighting its multilingual precision and real-time responsiveness.
The model’s debut coincides with Google’s broader Gemini Audio initiative, which aims to power natural dialogue systems. It replaces Chirp 3 in products like Docs, Keep, and Gmail, while also enabling speaker attribution for up to three voices in pre-recorded audio. Users can customize vocabulary to prevent misinterpretation of technical terms or names like 'Vedat Muriqi' or 'Burnley'.
The launch arrives amid intensifying competition in artificial intelligence. OpenAI is preparing for an IPO by end-2026, focusing ChatGPT on enterprise productivity tools to convert its 900 million weekly users into high-compute customers. Meanwhile, Salesforce and Anthropic unveiled Claudeforce, integrating Claude AI into Salesforce workflows, with an open beta scheduled for September 2026.
Google’s move reflects a strategic push to embed AI into everyday productivity. By improving voice-driven workflows, Gemini 3.5 Transcribe addresses llm training data requirements for nuanced speech understanding. However, the delay in releasing Gemini 3.5 Pro—promised in June—raises questions about Google’s roadmap amid OpenAI’s aggressive coding-focused Gemini 3.7 Flash rollout.
The market reaction underscores AI’s competitive urgency. While Gemini 3.5 Transcribe advances real-time transcription, its success will depend on seamless integration into Chrome and broader user adoption. Competitors like OpenAI and Anthropic are racing to match these capabilities, particularly in enterprise applications where productivity tools dominate.
FAQ
What is Gemini 3.5 Transcribe? It is Google’s latest speech-to-text model, offering 70% lower latency and 5.5% WER, designed for voice editing and multilingual support.
How does it improve over Chirp 3? It reduces transcription time, handles disfluencies, and supports 85+ languages with better contextual accuracy.
Where is it available? It powers Gboard Rambler, Gemini apps, and will expand to Chrome for voice-to-text in web fields.
What’s the competitive context? OpenAI focuses on ChatGPT productivity, while Salesforce-Anthropic’s Claudeforce targets enterprise workflows, intensifying the AI race.
