Speech-to-Speech Index
An evaluation benchmark designed to measure the overall capability of speech-to-speech models. It produces a weighted composite score across multiple dimensions, including speech comprehension accuracy, response relevance, naturalness rating, and time-to-first-token latency. The benchmark is maintained by Google.
Timeline (last 90 days)
Gemini 3.8 Live's Speech-to-Speech Index score is 76.0, with a first audio output latency of only 1.18 seconds.
Speech-to-Speech Index is an evaluation benchmark measuring the comprehensive capabilities of speech-to-speech models, encompassing a weighted score across multiple dimensions including speech comprehension accuracy, response relevance, naturalness rating, and first-token output latency.
Gemini 3.8 Live's Speech-to-Speech Index is 76% for the standard version, with the Extended Thinking tier improving it to 82.6%
The Extended Thinking version topped the overall leaderboard on Artificial Analysis's speech-to-speech quality index with a score of 82.6.