Google DeepMind released Gemini 3.8 Live and 3.8 Live Extended Thinking, two new real-time audio models that directly challenge OpenAI's recently launched GPT-Live-1. The pricing gap is substantial. Google's offering costs $1.38 per hour of voice conversation, undercutting OpenAI's pricing by a significant margin.
Both models top the Artificial Analysis speech-to-speech leaderboard, positioning Google as the performance leader in real-time voice AI. The Extended Thinking variant adds reasoning capabilities to the mix, allowing the model to deliberate before responding to voice queries. This feature appeals to developers building applications that require more complex problem-solving in conversational settings.
The real-time voice market has heated up rapidly. OpenAI introduced GPT-Live-1 earlier this year with a focus on naturalness through full-duplex audio technology. Full-duplex allows simultaneous speaking and listening, closely mimicking human conversation without the awkward pauses of turn-based interaction. OpenAI's implementation remains technically superior in this regard. GPT-Live-1 delivers a more natural conversational experience because users can interrupt and overlap speech naturally.
Google's Gemini 3.8 Live models achieve better performance on standardized benchmarks while operating at a lower cost. This creates a classic trade-off scenario for developers. Those prioritizing performance per dollar will favor Gemini. Those building applications where conversational naturalness is non-negotiable may still choose GPT-Live-1 despite higher costs.
The Extended Thinking variant adds strategic value. This model type takes longer to respond but produces more reasoned outputs. Developers using it gain a reasoning layer without switching models. Applications like customer support, technical assistance, and educational tools benefit from this capability. The cost remains competitive even with the added thinking overhead.
Pricing power matters in AI markets. Google's aggressive pricing reflects its infrastructure advantage and manufacturing scale. The company controls chip production through custom tensor processors, lowering per-unit inference costs. OpenAI lacks this vertical integration, relying on third-party hardware providers. This structural difference translates directly to customer pricing.
The leaderboard positioning matters too. Artificial Analysis measures speech-to-speech quality objectively. Topping these benchmarks gives Google credibility with enterprise customers evaluating options. Developers can point to independent measurements when justifying platform choices internally.
Market dynamics will shift as both companies iterate. OpenAI could improve full-duplex implementation or reduce pricing. Google could prioritize conversational naturalness in future releases. Neither company rests on current positions. The real-time voice space remains early stage with plenty of room for improvement.
Developer adoption will determine winners. Google offers the Gemini models through its AI Studio and Vertex AI platforms, making integration straightforward for teams already using Google Cloud. OpenAI provides GPT-Live-1 through its API with similar accessibility. Both lower barriers to experimentation.
The market implication is clear. Real-time voice AI moves from research curiosity to production reality. Applications that previously required complex audio processing pipelines now route through single API calls. Developers can build voice interfaces faster and cheaper than ever before. This acceleration will spawn new categories of voice-first applications across customer service, education, healthcare, and entertainment.
Google's entry at lower cost with matching performance benchmarks signals confidence in its technical direction. The company is playing a long game, building ecosystem lock-in through aggressive pricing and platform integration. OpenAI maintains quality advantages that justify premium positioning. Market segmentation emerges, with different winners in different use cases.