Key points
- ElevenLabs released two speech-to-speech models, Eleven Realtime Max and v1 Mini, through its ElevenAgents platform and an OpenAI-compatible Realtime API.
- The company says no model on the Artificial Analysis leaderboard beats Max on both speed and accuracy; Max reports 98.8 percent on Big Bench Audio.
- Eleven Realtime Translation, a live voice translation model, is in alpha for ElevenAPI enterprise customers and covers more than 90 languages.
- Deutsche Telekom is using the translation model to build calling features for its Magenta AI Call Assistant.
Why it matters for marketers: Voice agents and multilingual support get easier to prototype, and brands can keep one consistent voice across markets, though the translation model is enterprise-only alpha for now.
ElevenLabs released two speech-to-speech models under the name Eleven Realtime. The larger Max version is aimed at harder jobs such as customer support, while the smaller v1 Mini targets natural, lower-cost conversation and is built to tell a finished turn from a passing acknowledgment like "mm-hmm." Both are available on ElevenAgents and through an OpenAI-compatible Realtime API.
On performance, ElevenLabs says Max reaches 98.8 percent accuracy on the Big Bench Audio test, with responses in just over half a second. The company claims no model on the Artificial Analysis leaderboard currently combines better speed and accuracy. Those are vendor claims, and teams should check them against their own call scenarios.
Separately, ElevenLabs announced Eleven Realtime Translation, which lets two people speak different languages and hear each other in their own during a live call. It is in alpha for ElevenAPI enterprise customers. The company points to telecom networks, contact centers and live events as use cases, and says Deutsche Telekom is already building features on it for its Magenta AI Call Assistant.
For marketers and content teams, the draw is consistency. The translation model can hold one chosen voice, from a library or a clone, across more than 90 languages, so a brand or agent sounds the same in every market. Teams can also trade translation quality against speed depending on the call type.
Next, watch whether access widens beyond enterprise alpha and how pricing compares, since the Realtime models are billed per minute plus tokens, with Max costing more than Mini. Customer-facing teams should test latency and voice quality in their key languages before committing.
Sources
This brief is our own summary. Read the full reporting at the source.
