OpenAI
9 models benchmarked on 3 benchmarks. Pick one to see how it did.
Last updated Methodology by Dileep Chagam
OpenAI
- Speech to speech3
- GPT Realtime 2.179.3% reliable, 2nd of 1279.3% reliable, 2nd of 12
- GPT-Live 1 + GPT-6 Sol75.6% reliable, 3rd of 1275.6% reliable, 3rd of 12
- GPT Realtime 2.1 Mini50.0% reliable, 11th of 1250.0% reliable, 11th of 12
- Turn Bench3
- GPT-Live 185.5% turn-taking, 4th of 1385.5% turn-taking, 4th of 13
- GPT Realtime 2.184.6% turn-taking, 5th of 1384.6% turn-taking, 5th of 13
- GPT Realtime 2.1 Mini77.4% turn-taking, 12th of 1377.4% turn-taking, 12th of 13
- Speech to text3
- 4o Mini Transcribe3.35% words wrong, 13th of 193.35% words wrong, 13th of 19
- 4o Transcribe3.90% words wrong, 16th of 193.90% words wrong, 16th of 19
- Realtime Whisper2.16% words wrong, 7th of 192.16% words wrong, 7th of 19