Google AI Launches Gemini 3.5 Transcribe Model

Google AI has released Gemini 3.5 Transcribe, a new speech-to-text model designed to handle multilingual audio processing. According to MarkTechPost, the model reports an average word error rate of 2.6 percent across a broad spectrum of more than 85 languages.

The system aims to improve transcription accuracy and lower error rates for developers building applications that require real-time or batch audio processing. The release targets high-performance speech recognition needs across diverse linguistic datasets, offering developers improved accuracy metrics compared to previous iterations.

Based on reporting by www.marktechpost.com.