Google's Gemini 3.8 Leaps Ahead in AI Voice Performance

Google's latest AI voice model, Gemini 3.8 Live Extended Thinking, has outperformed its competitors in a key benchmark, according to a recent evaluation by Artificial Analysis.

The model scored 82.6 on the Speech-to-Speech Index, surpassing models from OpenAI and xAI, indicating significant improvements in AI conversations.

In addition to its performance gains, Gemini 3.8 Live Extended Thinking has achieved cost reductions, with a standard price of $0.84 per hour of input audio, less than half the rate of its predecessor, which cost $1.75 per hour.

However, Google's older model, Gemini 3.1 Flash Live, still excelled in blind testing when human listeners were involved, with the new standard model ranking second and Extended Thinking trailing despite completing more tasks.

This development has significant implications for the voice AI market, where Alphabet Inc., Google's parent company, is racing to improve user experience and outperform its competitors.