ElevenLabs launched Eleven Realtime on 20 October, which it calls “a new class of Speech to Speech models.” Instead of wiring speech recognition, turn-taking, an LLM and text to speech together yourself, each model is a complete voice agent. Both support more than 90 languages and any of the 17,000+ voices in the ElevenLabs library.
Two models
- Eleven Realtime v1 Mini is the low-cost option for assistants, tutors and interactive characters. ElevenLabs says it “will be sufficient for most use cases.”
- Eleven Realtime v1 Max targets support, lead qualification and multi-step workflows where the agent follows a process and calls tools.
The numbers ElevenLabs reports
ElevenLabs says Artificial Analysis evaluated both models on three public benchmarks. As published in the launch post:
| v1 Mini | v1 Max | |
|---|---|---|
| Big Bench Audio (speech reasoning) | 75.9% | 98.8% |
| Average time before it starts speaking | 0.62 s | 0.57 s |
| Full-Duplex-Bench (pauses, interruptions, turn-taking) | 95.8% | 94.1% |
| Speech Agent Arena score | 988 | 998 |
| Tool-use task completion | 60.3% | 87.1% |
The tool-use gap is the one to read: ElevenLabs itself recommends Max for agents that call tools heavily.
How to use it
In ElevenAgents, you pick v1 Mini or v1 Max from the model dropdown on a new or existing agent; ElevenLabs says nothing changes for existing agents unless you update them. For developers, the Realtime API follows OpenAI’s Realtime spec over WebSocket at wss://api.elevenlabs.io/v1/convai/realtime, with model names eleven_realtime_v1_mini and eleven_realtime_v1_max. The launch post shows OpenAI’s @openai/agents-realtime SDK pointed at that URL. Two limits: the API is WebSocket only, so a WebRTC integration has to move first, and it runs separately from ElevenAgents, so channels and the knowledge base need the ElevenAgents platform.
Price
The API pricing page lists both models at $0.08 per minute of connected audio plus LLM usage. Token rates per million are $0.20 input and $6.00 output for Mini, and $1.80 input and $9.00 output for Max. The included minutes on each plan are ElevenLabs’ estimates from typical token use.
It arrived the same day as Mistral’s Voxtral 3, which takes the opposite route: separate transcription, LLM and speech models joined in one streaming pipeline. If you are choosing, the trade is control over each stage against a single model you tune as a whole.
The Campfire
No commentsNobody has pulled up a log by this one yet. Be the first to say what you make of it.
Held for the desk. It appears after a look.