Voice AI startup Phonic says its new model can answer a caller more than 14 seconds faster than an OpenAI rival in a side-by-side demo. The San Francisco company also announced a $27.5 million Series A round on October 29, 2026.
The round was led by First Round Capital. Lux Capital, Felicis, Definition and Essence VC also took part, bringing Phonic’s total funding to $31.5 million.
The new model, called Phonic-1, is a speech-to-speech system built for voice agents that handle tasks over the phone. The company said it ranked first on the Artificial Analysis Speech to Speech Index at launch.
How the Speed Test Worked
In a demo posted by Phonic, the model made five tool calls while a caller was still speaking. Tool calls are actions such as looking up a flight or checking a booking.
Phonic-1 replied 1.1 seconds after the caller stopped talking. OpenAI’s GPT-Live made all five tool calls only after the caller finished, and it replied 15.7 seconds later.
Phonic says its model is the only one that can act while the caller is still talking. Other models typically wait, tell the caller they will look something up, and then call a tool.
Most voice agents today use a chain of three models. One turns speech into text, a second writes a reply, and a third reads that reply aloud.
Phonic argues that this chain loses details found in a real voice. These include frustration, long pauses and the way a person says their own name. Phonic-1 handles speech directly inside one model.
Humans Still Sound More Natural
A third party, Podonos, ran blind listening tests comparing Phonic-1 with other voice models. Judges preferred Phonic-1 over OpenAI’s GPT-Live-1 by 61.5% to 32.0%.
Phonic-1 was also preferred over Grok Voice 3 by 66.5% to 26.5%. It beat ElevenLabs’ Eleven v4 Turbo by 78.0% to 14.0%.
Human voice actors still came out ahead. Listeners chose human voices over Phonic-1 by 57.0% to 32.0%, and the company said there is “substantial room to improve” on naturalness.
On multilingual tasks, Phonic-1 scored 83.2 on Sierra’s τ³-Voice benchmark. OpenAI’s GPT-Live-1 scored 79.9 and Google’s Gemini 3.8 Live scored 68.5.
The model was trained on more than 50 languages. It can switch languages mid-call while keeping the same voice. One demo showed a 44-minute call to a hotel in Brazil that included 123 tool calls.
Phonic says its technology now powers calls on six continents. Customers include Ashby, Maven AGI, Rho and a Fortune 50 IT infrastructure company.
Ashby co-founder Abhik Pramanik said Phonic powers the hiring platform’s AI interviews in 25 countries. Angel backers include Naval Ravikant, Hugging Face’s Clement Delangue and Replit’s Amjad Masad.
The funding comes as larger firms move into voice AI. Salesforce has agreed to acquire Cresta, and Amazon offers its own speech-to-speech model called Nova Sonic.
Phonic-1 is available now. The company is offering $250 in free credits to users who sign up by November 30, and three free months to customers who switch from other providers.
