The Pulse
Phonic launches Phonic-1 and raises $27.5M for speech AI
Phonic introduced its speech-to-speech model, Phonic-1, alongside a $27.5 million Series A led by First Round. The company says the model can call tools while a caller is still speaking and is available now.
AI.info Team · Written with AI · fact-checked ·
“The speech-to-speech model was ready for production, and its voices are some of the most natural we’ve heard.”
Abhik Pramanik, Co-founder & VP of Engineering, Ashby
Phonic introduced Phonic-1 on October 29, pitching it as a speech-to-speech model that can start tool calls before a caller finishes speaking. The launch came with news of a $27.5 million Series A led by First Round. The company says the financing brings its total funding to $31.5 million.
The product targets voice agents handling tasks such as customer support, recruiting and healthcare calls. Rather than transcribing speech, generating a text response and converting it back to audio, Phonic says its model handles conversation directly as speech. The company argues that this lets the system retain cues such as tone and pauses while it reasons and responds.
Phonic-1 starts work mid-sentence
Phonic’s central product claim is that the model can launch a tool call while a caller is still speaking, rather than waiting for the end of the turn. The launch page’s audio-demo caption says Phonic-1 calls all five tools while the caller is still talking and replies 1.1 seconds after they stop; GPT-Live calls the tools after the caller stops and replies after 15.7 seconds.
The company also presented a 44-minute hotel-booking call in Brazil. In the demonstration, a maid of honor asks a hotel for help booking a bachelorette party, and Phonic-1 makes 123 tool calls. The demonstration is a showcase, not a measure of typical call performance. It illustrates the kind of multi-step interaction Phonic is targeting: a voice agent that can gather information and complete actions without forcing the caller through long pauses.
A top spot on Artificial Analysis’ index
Phonic says Phonic-1 took the number one spot on Artificial Analysis’ Speech to Speech Index. Artificial Analysis describes the index as a weighted average of speech reasoning, agentic performance, arena preference and task success rate. The index is a composite benchmark, not a single measure of how a model performs in every live conversation.
Phonic’s launch post also reports a score of 83.2 on Sierra’s multilingual full-duplex benchmark, ahead of GPT-Live-1 at 79.9 and Gemini 3.8 Live at 68.5. The company says Phonic-1 is trained on more than 50 languages and can switch languages mid-conversation while keeping the same speaker identity.
Naturalness remains a stated weakness
Phonic says third-party blind listening tests found that participants preferred Phonic-1 a majority of the time when comparing model responses. The company also compared the model with human voices and says humans won roughly 60% of the time. It describes naturalness as an area for further improvement.
That caveat matters for businesses weighing whether task completion and speed are enough to offset a voice that listeners may still find less convincing than a person. Phonic’s announcement presents the model’s ability to act during a caller’s turn as a central advantage, while acknowledging that its naturalness remains a work in progress.
$27.5 million backs production expansion
First Round led the Series A, with Lux Capital, Felicis, Definition and Essence VC also participating. Phonic says its systems now handle calls across six continents for customers including Ashby, Maven AGI, Rho and a Fortune 50 IT infrastructure company. Ashby’s Pramanik said the system powers the company’s AI interviews in 25 countries.
Phonic-1 is available now, and the company offers $250 in free credits to people who sign up by November 30. Phonic also offers three months at no charge to customers switching from existing voice systems, according to the launch post. The company’s comparison with human voices points to the remaining naturalness gap even as it promotes the model’s ability to take action in real time.