Voice mode is the dimension that most exposes weak AI companion infrastructure. Latency, prosody, and the ability to interrupt cleanly separate the platforms below from the broader category. Each entry was tested for at least three live voice sessions exceeding fifteen minutes.
8.8
8.5
7.0
6.4
Every entry above was independently tested by our editorial team. The scoring rubric, including weighting between memory, persona, pricing, voice, and image dimensions, is published in full in our methodology document. Read the full methodology.