Gemini 3.5 Transcribe now available on AI Gateway
Vercel’s AI Gateway now supports Google’s Gemini 3.5 Transcribe, offering real-time and batch audio-to-text transcription with language detection and custom vocabulary support.
Vercel has integrated Google’s Gemini 3.5 Transcribe into its AI Gateway, enabling developers to convert audio recordings into text through two methods. The batch model processes entire files in a single request, while the live variant streams transcriptions in real time via WebSocket, updating as audio is received. Both variants automatically detect over 85 languages and can follow speakers who switch languages mid-conversation.
The live transcription feature accepts raw audio chunks via a ReadableStream, allowing direct microphone input without intermediate processing. Developers can specify the audio format using inputAudioFormat, while the batch model handles complete recordings. For testing without code, users can access a browser-based demo to send audio and receive transcripts in real time.
AI Gateway serves as a unified interface for model access, usage tracking, and cost management, with built-in failover and performance optimizations to improve reliability beyond provider baselines. It includes features like custom reporting, API key budgeting, and routing rules to streamline workflows and enforce governance across teams.
Pricing for Gemini 3.5 Transcribe on AI Gateway aligns with Google’s provider rates, with no additional platform fees applied to inference costs, including Bring Your Own Key requests. Users can explore all available transcription models or follow Vercel’s speech quickstart guide to begin implementation.