Vercel adds 14 ElevenLabs audio models to AI Gateway

Vercel and ElevenLabs are partnering to bring 14 ElevenLabs models to AI Gateway. The models span six audio types: speech to text, streaming text to speech, realtime voice agents, music, sound effects and multi-voice dialogue. All of them are reached through a single gateway API key, and requests show up in the same usage, cost and trace views as other AI Gateway traffic.

Vercel's example pipeline transcribes microphone audio with elevenlabs/scribe-v2-realtime, sends the transcript to a language model, and streams that text into elevenlabs/eleven-flash-v2.5 for speech. Because each step is a separate model, the transcription or speech model can be changed without touching the rest of the pipeline.

For conversations that need back-and-forth, elevenlabs/eleven-realtime-v1-max and elevenlabs/eleven-realtime-v1-mini handle listening and speaking in one model over a single WebSocket session. The post also shows a gateway token call that lets a browser open that session directly.

Agent builders can call ElevenLabs voice, transcription and music models through the gateway key they already use, instead of wiring up a separate integration. Choose the chained transcribe, reason and speak pipeline when you want control over each step, or a single realtime model when latency and natural turn-taking matter more. The 50% discount ends November 9, so plan costs for longer use at regular prices.