On May 8, OpenAI added GPT-Realtime-Translate to its Realtime API, the first OpenAI model purpose-built for live speech translation. The launch prompted a question event producers have been asking us all week: how does the new OpenAI model compare to the live translation systems that already serve real events?
We ran it through the same head-to-head evaluation harness we use on our own product and three of the most-cited competitors: Google Meet, LiveVoice, and Palabra. The short version:
- VoiceFrom Interpret is substantially more accurate than OpenAI’s new model on the industry-standard WMT-grade quality metric, on average across eight language pairs.
- OpenAI’s median latency is faster than VoiceFrom’s.
- Full data, scoring methodology, charts, and audio comparisons are in the engineering writeup below.
Full benchmark: Five platforms, one harness: a head-to-head live translation benchmark.
If accuracy is what your audience needs, VoiceFrom is the system to put on stage. Talk to us and we’ll set up a pilot at your next event.
