Elevenlabs launches Eleven Realtime: 98.8% at 0.57 seconds

Elevenlabs' pitch rests on a single pairing: high accuracy and low latency. The v1 Max's 98.8 percent on Big Bench Audio and 0.57-second response time are presented together, and the company says no model on the Artificial Analysis leaderboard is both faster and more accurate right now. That claim is the company's own, and it is attributed to Elevenlabs rather than an independent evaluation.

The two variants split the market by task. V1 Mini is framed around the mechanics of natural conversation — detecting when someone has finished speaking, or is only offering a filler like "mm-hmm" — at lower cost. V1 Max is positioned for more complex work such as customer support, where errors are costlier and the per-minute price is correspondingly higher than v1 Mini's.

Distribution follows the pattern Elevenlabs has already established: the models are available through its ElevenAgents platform and through an OpenAI-compatible Realtime API, which lowers the switching cost for developers already building against OpenAI's realtime interface. Pricing is per minute plus token costs, so buyers will need to weigh latency gains against a bill that scales with conversation length.