Today, we’re introducing Sunlight 1.0 Realtime, our first model built specifically for voice agents and real-time conversations.
Sunlight 1.0 Realtime marks an important expansion of the Sunlight model family. Until now, our models have primarily focused on text-based intelligence and general-purpose AI experiences. Realtime is designed around a different challenge: enabling AI systems to participate in conversations as they happen.
That means prioritizing low latency, natural turn-taking, fast responses, and continuous interaction.
Built for real-time voice experiences
Voice interfaces require a fundamentally different interaction model from traditional text-based AI.
People naturally interrupt, pause, change direction mid-sentence, and expect responses immediately. Sunlight 1.0 Realtime was developed with these behaviors in mind, giving developers a foundation for building voice experiences that feel significantly more responsive and conversational.
The model is designed for applications including:
- Voice assistants
- Customer support agents
- AI companions
- Interactive voice applications
- Real-time conversational interfaces
- Voice-enabled productivity tools
- Embedded AI agents
Rather than treating voice as an additional layer on top of a text model, Realtime is designed around the requirements of live interaction from the beginning.
Faster, more natural conversations
One of our primary goals with Sunlight 1.0 Realtime is reducing the delay between a person speaking and an AI responding.
Small delays can dramatically change how natural a conversation feels. Realtime focuses on maintaining fluid exchanges where users can speak naturally instead of adapting their behavior to the model.
This also creates new opportunities for developers building agents that need to react quickly to changing context during a conversation.
The beginning of Sunlight Realtime
Sunlight 1.0 Realtime is our first release in this category.
We expect the Realtime family to evolve significantly as we improve conversational intelligence, latency, voice understanding, interruption handling, tool use, and agent capabilities.
This release establishes the foundation for that work.
Over time, we want Sunlight models to support AI experiences that can communicate and act with the immediacy people expect from a real conversation.
API availability
Sunlight 1.0 Realtime is being released today, with API access becoming available on October 9, 2026, one week after the initial release.
The additional rollout period will allow us to prepare infrastructure, monitor the initial deployment, and ensure the API can support developers building latency-sensitive applications at scale.
Once available, developers will be able to integrate Sunlight 1.0 Realtime directly into their own applications and voice-agent infrastructure through the Sunlight API.
Additional API documentation, pricing, limits, and implementation guidance will be published alongside API availability.
What comes next
Realtime conversations are becoming an increasingly important part of how people interact with AI.
Sunlight 1.0 Realtime is our first step toward making those interactions faster, more natural, and more useful across a much broader range of products.
And it’s only the beginning.
