Elevenlabs has released two new speech-to-speech models called Eleven Realtime, designed specifically for real-time artificial intelligence conversations. The larger variant, known as Eleven Realtime v1 Max, achieved a score of 98.8 percent on the Big Bench Audio benchmark while maintaining a rapid response time of 0.57 seconds. According to Elevenlabs, no other model currently listed on the Artificial Analysis leaderboard matches this combination of speed and accuracy.
The lineup also includes a smaller model, v1 Mini, which is engineered for natural conversations at a lower cost and features the ability to detect when a user has finished speaking rather than just pausing. While the v1 Mini focuses on efficiency, the v1 Max targets more complex conversational tasks such as automated customer support. Both models are integrated into the ElevenAgents platform and are accessible through an OpenAI-compatible Realtime API, with pricing structured per minute alongside token costs.
These new voice models matter because they significantly lower the latency and improve the accuracy of real-time conversational artificial intelligence applications.