ElevenLabs has launched two real-time voice models for businesses, with Mini prioritising natural conversation and Max targeting multilingual support agents that can use tools and complete multi-step tasks.
ElevenLabs has expanded its real-time voice offering with two new models aimed at businesses building conversational assistants and automated support systems. The company unveiled Eleven Realtime v1 Mini and Eleven Realtime v1 Max on 21 October 2026, saying both are built to process speech and generate spoken replies within the same system rather than relying on separate transcription, reasoning and text-to-speech components.
The launch strengthens ElevenLabs’ push into enterprise voice automation, a market in which firms are increasingly looking for tools that can handle live customer interactions in multiple languages. According to the company’s materials, the new models support more than 90 languages and can draw from a voice library of over 17,000 options, giving developers a wide range of deployment choices for support desks, interactive assistants and character-driven applications.
The two models are tuned for different jobs. Mini is pitched as a lower-cost option for natural conversation, while Max is aimed at more demanding workflows in which an agent must follow instructions, call external tools and complete multi-step tasks. In benchmarking released by ElevenLabs, Max substantially outperformed Mini on Big Bench Audio, a spoken reasoning test, and also posted stronger results on tool-use tasks, suggesting it is better suited to customer service systems that need to do more than simply talk. Mini, however, did slightly better in a benchmark focused on natural conversation, where turn-taking, interruptions and short acknowledgements matter more than task completion.
That split matters because voice automation is not a single capability. Faster response times, conversational fluency and reliable execution of procedures are related but distinct requirements. ElevenLabs’ own testing showed Max speaking marginally faster than Mini, while the company’s conversational benchmarks pointed to Mini having an edge in dialogue flow. For businesses, that means the right model is likely to depend on whether the priority is a smoother sounding exchange or a more capable agent that can work through structured support processes.
ElevenLabs is also trying to reduce the integration burden for developers. The company says its interface is compatible with OpenAI’s Realtime API, which should allow existing code to be adapted by changing the connection address and model name. Even so, the current setup supports WebSocket connections only, which means teams relying on WebRTC would need to adjust their communication stack before deploying the new models. Independent reviews of ElevenLabs’ broader platform suggest the company has invested heavily in documentation and developer support, with detailed API guides, code samples and tutorials intended to make adoption easier.
The release fits into a wider product strategy that has turned ElevenLabs into more than a voice synthesis vendor. External reviews of its agents platform describe a multilingual support product that can integrate with systems such as Zendesk, Salesforce and ServiceNow, while the company’s own materials position its offering as a broader voice infrastructure stack spanning text-to-speech, speech-to-text, voice cloning and conversational agents. Taken together, the new models appear designed to push that platform deeper into live customer service, where businesses want both natural sounding conversation and dependable task execution.
Disclaimer: This content is intended for informational purposes only. Readers are advised to exercise their own judgement, conduct due diligence, or consult a qualified expert before acting on any information provided.
