Smallest.ai has raised $13 million in Series A funding to build voice agents that respond quickly enough to feel more like a natural conversation.
Seligman Ventures led the round, with Sierra Ventures and 3one4 Capital also participating. The investment brings the startup's total funding to more than $21 million since its founding in late 2024.
A smaller model for the live conversation
Most voice agents convert speech, wait for a large language model to process a complete request, and then generate an answer. Even a short pause can feel unnatural during a phone call. Smallest.ai is instead developing a specialized model that listens, thinks, and prepares a response while the speaker is still talking.
The system is designed for constrained customer-service topics where speed matters more than broad knowledge. When a conversation moves outside that smaller model's expertise, it can hand the question to a larger model and briefly place the caller on hold while it works on the answer.
CEO Sudarshan Kamath argues that this two-model pattern could become standard: a compact real-time system handles the conversation, while a larger offline model solves difficult requests when needed. That division could reduce latency and infrastructure costs without giving up access to more capable reasoning.
Smallest.ai is also focusing on accents, multiple languages, noisy environments, interruptions, and other details that make voice interaction harder than text chat. Existing customers include RingCentral and Truecaller. Competitors include ElevenLabs, Cartesia, and regional voice-AI developers such as Sarvam.
More human-sounding systems also create disclosure and consent questions. Companies will need to make clear when callers are speaking with software, even if the voice and timing become difficult to distinguish from a person.
