Google DeepMind Launches Gemini 3.8 Live Models, Undercutting OpenAI on Price for Voice Agents

Google DeepMind released Gemini 3.8 Live and Gemini 3.8 Live Extended Thinking on September 15, 2026, introducing new speech-to-speech models for production voice agents that significantly undercut OpenAI's GPT-Live-1 on price.

New Gemini 3.8 Live Models Detailed

The Gemini 3.8 Live models are engineered for distinct use cases. Gemini 3.8 Live focuses on scalability and cost efficiency, making it suitable for high-volume applications. In contrast, Gemini 3.8 Live Extended Thinking is tailored for complex, multi-step reasoning tasks, addressing scenarios that require deeper analytical capabilities from a voice agent.

These models enable voice agents to process visual input in near real-time, execute tool and API calls in the background, and seamlessly switch between 97 supported languages during a conversation. This multimodal and multilingual functionality aims to enhance the versatility of AI-powered voice interactions.

Availability and Integration

Google has made the Gemini 3.8 Live models accessible through multiple channels:

• Developers: Available via the Gemini API and Google AI Studio.
• Enterprises: Offered in private preview through Gemini Enterprise.
• Consumers: Rolling out to users in Search Live.

At launch, ecosystem partners include Salesforce, Genspark, and Lumeris. Developer integrations are supported by platforms such as Agora, LiveKit, LangChain, Pipecat, and Vercel, facilitating broader adoption and application development.

Competitive Pricing Against OpenAI's GPT-Live-1

A key aspect of Google DeepMind's new offering is its competitive pricing structure. Google charges $0.005 per minute for audio input and $0.018 per minute for audio output. This translates to an approximate cost of $1.38 for an hour of voice conversation.

In comparison, OpenAI's GPT-Live-1 is priced at a minimum of $0.05 per minute, totaling at least $3.00 per hour for voice conversations. This positions Google's Gemini 3.8 Live models as a more cost-effective option for developers and businesses utilizing conversational AI.

Performance and Quality Benchmarks

Gemini 3.8 Live Extended Thinking has achieved the top rank on Artificial Analysis' Speech-to-Speech Quality Index, scoring 82.6. This benchmark indicates its strong performance in speech-to-speech capabilities.

However, The Decoder suggests a distinction in conversational quality, noting that OpenAI's GPT-Live-1 may offer more natural conversations due to its full-duplex audio capabilities, while Google's models appear to prioritize price optimization.

Security and Transparency Features

All audio generated by Google's Gemini 3.8 Live models incorporates a SynthID watermark. This feature is designed to enhance transparency by allowing identification of AI-generated content, addressing concerns related to synthetic media.

Conclusion

Google DeepMind's introduction of Gemini 3.8 Live and Gemini 3.8 Live Extended Thinking marks a significant development in the landscape of production voice agents. With a focus on cost efficiency, advanced reasoning, and broad accessibility, these models aim to expand the capabilities and adoption of AI-powered conversational technologies. The competitive pricing strategy directly challenges existing offerings, particularly OpenAI's GPT-Live-1, while the integration of features like visual input processing, tool execution, and multilingual support positions them for diverse applications. Developers and enterprises should evaluate these new models for their specific needs, considering both their performance benchmarks and cost advantages.

Sources

• Gemini 3.8 Audio (Live, Live Extended Thinking) - Model Card
• Introducing Gemini 3.8 Live and 3.8 Live Extended Thinking
• Introducing Gemini 3.8 Flash and 3.8 Flash Cyber
• Gemini 3.8 Live & Gemini 3.8 Live Extended Thinking
• Google launches Gemini 3.8 Live to take on OpenAI's GPT-Live-1 at a fraction of the cost