Elevenlabs released two Eleven Realtime speech-to-speech models for real-time conversations with AI, The Decoder reports. The v1 Max variant, according to the publication, scored 98,8% in the Big Bench Audio test and showed a response time of 0,57 seconds.
The company also states that no model in the Artificial Analysis ranking combines higher speed and accuracy simultaneously. This claim in the available material is attributed to Elevenlabs' position, not an independent verification.
For voice AI users, latency and quality are interconnected constraints: a fast reply reduces pauses in the dialogue, but it does not by itself prove greater practical usefulness. The full testing conditions, the list of the second model, and independent confirmation of results are not provided in the original package.
editorial commentary
Why it matters
If the claimed metrics are confirmed by independent tests, the main consequence will be intensified competition for fast voice interfaces. The next observable signal will be publication of methodology and results of third-party comparisons. A significant uncertainty is that only the metadata summary from The Decoder is currently available.