Smallest.ai raises $13M to build ultra-fast voice AI that sounds genuinely human
7:47 AM PDT · July 31, 2026
Smallest.ai, founded in late 2024, is betting the next leap in voice agents will come from smaller, specialized models built for human conversation — making speaking to an AI agent indistinguishable from talking to a human.
Funding
21 million.
Product thesis
CEO/founder Sudarshan Kamath: humans listen, think, and speak simultaneously and interrupt — the model is designed to work the same way. Standard LLMs take a full prompt then think; in voice, even short pauses feel unnatural.
Real-time intelligence layer for natural customer conversations on specific topics with virtually zero response lag. Out-of-knowledge queries hand off to a large foundational model (brief “research” hold). Kamath believes agents will use two models: small voice model for real-time interaction + offline LLM for complex problems.
Focus on voice-specific nuances: accents, dozens of languages, noisy environments. Customers include RingCentral and Truecaller; potential customers include customer-support AI firms (Sierra, Decagon). Competes with ElevenLabs, Cartesia, and regional players like Sarvam. Strict focus on real-time conversational voice agents for enterprise (not dubbing/podcasting).
Goal stated: “break the Turing test” — callers should not know whether they are speaking to AI or human.