Glonce

Elevenlabs releases Eleven Realtime, two speech-to-speech models for live conversations

modelsThe Glonce editorsThe Decoder🔥65Glonce rating65Impact55Novelty55Relevance100Surprise35Players60Credibility75
Elevenlabs releases Eleven Realtime, two speech-to-speech models for live conversations

Elevenlabs has released two new speech-to-speech models under the name Eleven Realtime, built for real-time AI conversations. The two variants are v1 Mini and v1 Max.

The headline numbers belong to the Max variant. It scores 98.8 percent on the Big Bench Audio benchmark and responds in 0.57 seconds. Elevenlabs says that, right now, no model on the Artificial Analysis leaderboard is both faster and more accurate than Max.

The two models are aimed at different jobs. The smaller v1 Mini is built for natural conversations at lower cost, and it can tell when someone is done talking and when they are just saying "mm-hmm." V1 Max targets more complex tasks such as customer support.

Both models are available through the ElevenAgents platform and through an OpenAI-compatible Realtime API. Pricing is per minute plus token costs, and v1 Max costs more than v1 Mini. The article does not give actual prices; it points readers to the Elevenlabs pricing page.

Key facts

  • Elevenlabs released two speech-to-speech models called Eleven Realtime: v1 Mini and v1 Max.
  • Elevenlabs reports that v1 Max scores 98.8 percent on the Big Bench Audio benchmark with a 0.57-second response time.
  • Elevenlabs says no model on the Artificial Analysis leaderboard is both faster and more accurate than Max right now.
  • v1 Mini targets lower-cost natural conversation, including telling the end of a turn from a backchannel like "mm-hmm"; v1 Max targets complex tasks such as customer support.
  • Both are available via the ElevenAgents platform and an OpenAI-compatible Realtime API, priced per minute plus token costs, with Max costing more.

Why it matters

Voice agents live or die on two things: how fast they answer and whether they follow what was said. Elevenlabs is claiming a lead on both at once for the Max variant, with 98.8 percent on Big Bench Audio and a 0.57-second response time. The Mini model's ability to tell a finished turn from a passing "mm-hmm" speaks to a different problem: not cutting people off or stalling mid-conversation.

Who it affects

Teams building voice agents on ElevenAgents or on the OpenAI-compatible Realtime API are the direct audience. v1 Max is pitched at more complex work such as customer support, while v1 Mini suits natural conversation where cost matters more.

How to use it

Both models are reachable through the ElevenAgents platform and an OpenAI-compatible Realtime API. Pricing is per minute plus token costs, with v1 Max costing more than v1 Mini. The exact rates are on the Elevenlabs pricing page. A reasonable split from the article's framing: Mini for lower-cost natural dialogue, Max for harder tasks like customer support.

How solid is it

The benchmark claim comes from Elevenlabs itself; no independent verification is mentioned. The 98.8 percent score and the 0.57-second response time are given for Max only, and the source does not say whether the response time is a mean, median or best case, or how it was measured. No scores or response times are given for v1 Mini.

Risks and caveats

The "fastest and most accurate" statement is Elevenlabs' own, tied to the Artificial Analysis leaderboard as it stands now, so it can change as other models are added. The article names no competitor models for comparison. It also gives no release date, no actual prices, and no details on supported languages, voices or model architecture.