Smallest.ai, founded in late 2024, has secured $13 million in a Series A financing led by Seligman Ventures, with participation from Sierra Ventures and 3one4 Capital. The new round brings the startup’s total funding to over $21 million.
The company’s thesis is that the next meaningful improvement in voice agents will come from smaller models designed specifically for human conversation rather than from making large language models faster. Smallest.ai aims to make talking to an AI agent indistinguishable from speaking with a human.
Listen, think, speak — concurrently
Smallest.ai is building a compact voice model that mirrors how humans listen, think and speak at the same time. Founder and CEO Sudarshan Kamath argues that standard LLMs wait to receive an entire prompt before they begin processing, which creates a latency profile acceptable in text chat but awkward in live voice interactions where even short pauses feel unnatural.
The startup’s model acts as a real‑time intelligence layer, enabling natural customer conversations on focused topics with virtually no response lag. When the small model encounters a topic beyond its limited knowledge base, Smallest.ai escalates the query to a larger foundational model and briefly places the customer on hold while the system “researches” the issue — mirroring how a human agent would behave.
A two‑model pattern and voice‑centric focus
Kamath expects most AI agents to adopt a dual‑model architecture: a small voice model for immediate interaction and an offline LLM called on for complex problems. Unlike broad foundational models, Smallest.ai concentrates on voice‑specific challenges such as handling diverse accents, supporting dozens of languages and operating in noisy environments.
Existing customers include voice‑centric companies such as RingCentral and Truecaller. Kamath said any customer support business is a potential client for Smallest.ai, including newer entrants. When asked why well‑funded customer support startups wouldn’t build their own voice models, he responded that becoming “extremely good at voice” would distract them from their core business.
Competition and positioning
Smallest.ai competes with voice AI providers like ElevenLabs and Cartesia, as well as regional players such as Sarvam that target local languages. While some competitors apply voice AI to use cases like audio dubbing and podcasting, Smallest.ai focuses exclusively on real‑time conversational voice agents for enterprise customers.
Kamath says the company’s single goal is to make models that pass the Turing test for spoken interaction: users should not be able to tell whether they are talking to AI or a human. The $13 million Series A will support further development of the real‑time voice models and expansion of the startup’s enterprise customer base.
