Smallest.ai has raised $13M in a Series A round to build voice models designed to make AI phone calls indistinguishable from human conversation.
Seligman Ventures led the round, with Sierra Ventures and 3one4 Capital participating. The startup, founded in late 2024, has now raised more than $21M.
The company’s pitch: the next leap in voice agents will not come from making large language models faster, but from smaller, specialized models built for the way humans actually talk, listening, thinking and speaking at the same time, with room for interruptions.
Kamath argues the block-style processing of large language models, tolerable in text chat, reads as robotic on a phone call, where even a beat of silence feels unnatural.
Smallest.ai’s model acts as a real-time intelligence layer for customer conversations with near-zero response lag. If a query falls outside its knowledge, it hands off to a large foundational model and briefly puts the caller on hold to research the issue, just as a human agent would.
Customers include RingCentral and Truecaller. The startup competes with ElevenLabs and Cartesia, but focuses strictly on real-time conversational voice for enterprises, avoiding dubbing and podcasting use cases.
Kamath’s stated goal is blunt: callers should not be able to tell whether they are talking to a person or to his company’s software.
