← Capa | Mercado

Google DeepMind's Gemini 3.8 Live tops speech-to-speech rankings at $1.38 per hour

2026-11-04 · Fonte:The Decoder

Google DeepMind's Gemini 3.8 Live tops speech-to-speech rankings at $1.38 per hour

Google DeepMind launches Gemini 3.8 Live and 3.8 Live Extended Thinking, claiming the Artificial Analysis speech-to-speech lead while undercutting OpenAI's GPT-Live-1 at $1.38 per hour of voice conversation.

Google DeepMind has put two new audio models into developers' hands: Gemini 3.8 Live and Gemini 3.8 Live Extended Thinking. The pair immediately claimed the top position on the Artificial Analysis speech-to-speech leaderboard, handing Google the leading spot in independent rankings for conversational voice AI. Both releases are aimed squarely at the developer market for spoken, voice-driven applications.

The headline numbers center on cost. Google prices the new models at $1.38 per hour of voice conversation — a rate that significantly undercuts OpenAI's GPT-Live-1 and works out to a fraction of its cost. For teams building voice agents or any service where spoken dialogue is billed by duration, the gap maps directly onto unit economics: the more conversation hours an application accumulates, the wider the absolute savings from routing that traffic to Google's stack.

Google's pricing advantage does come with a counterweight. OpenAI's GPT-Live-1, the report notes, should still sound more natural in conversation thanks to full duplex — the bidirectional audio capability that lets a voice system keep listening while it speaks. Naturalness of spoken interaction is difficult to capture in benchmark scores, which means procurement decisions may hinge on subjective listening tests rather than leaderboard placement alone. Google DeepMind, for its part, leads on the measured metric while also holding the price advantage.

For buyers evaluating voice infrastructure, the launch sharpens a trade-off that has now become explicit. Google DeepMind fields the top-ranked speech-to-speech models at $1.38 per conversation hour, while OpenAI counters with GPT-Live-1's full duplex naturalness at a higher cost. Which wins deployments will likely depend on whether a given application prioritizes billed minutes and benchmark standings, or the perceived fluidity and realism of spoken dialogue.

Tags: Google DeepMind · Gemini 3.8 Live · OpenAI · GPT-Live-1 · Speech-to-Speech · Voice AI


Leituras relacionadas


萬安算交所 AIXX · AIXX · Exchange verificado · Ir ao mercado