ElevenLabs launches Eleven Realtime voice AI models

ElevenLabs has introduced Eleven Realtime, a new line of AI models designed to power voice agents capable of natural, real-time interaction. The release includes two distinct versions: v1 Mini and v1 Max.

Eleven Realtime Mini is optimized for rapid conversations in applications such as virtual assistants, educational tools, and interactive characters. Eleven Realtime Max is engineered for complex, multi-step tasks, including customer service, lead qualification, and workflows requiring decision-making. In performance benchmarks by Artificial Analysis, the Max model achieved a 98.8% score on the Big Bench Audio reasoning test, significantly outperforming the Mini model's 75.9%.

The models support over 90 languages and access to more than 17,000 voices. Developers can utilize these models through the ElevenAgents platform—which supports channels like web, phone, and WhatsApp—or via the ElevenLabs Realtime API. While the managed implementation reduces the technical burden of handling turn-taking and orchestration, it limits a developer's ability to independently configure the underlying speech and reasoning components, a level of control only available in the company's Custom mode.