Phonic Launches Phonic-1, a Voice Model That Calls Tools While You're Still Talking, and Raises $27.5M
The startup's speech-to-speech model debuts at No. 1 on Artificial Analysis's index, narrowly ahead of xAI's Grok Voice 3.0, and starts at $0.07 a minute.
Published
Last updated / Fact-checked on
Voice AI startup Phonic on October 29, 2026, launched Phonic-1, a speech-to-speech model that it says is the only one able to call tools while a caller is still talking. The company also announced a $27.5 million Series A led by First Round. Phonic-1 debuted at No. 1 on the Artificial Analysis Speech to Speech Index, edging out xAI's Grok Voice 3.0, and its API starts at $0.07 a minute.
Image: Illustrative photo of a person on a phone call. Ehimetalor Unuabona via Unsplash / Wikimedia Commons — CC0 1.0
What Phonic-1 does differently
Most voice agents chain three models: one transcribes the caller, one decides what to say and one speaks. Phonic argues that a transcript drops tone, pauses and hesitation. Speech-to-speech models work on audio directly, but the company says they have traditionally traded intelligence for naturalness.
Phonic's headline claim is about timing. "Other models wait for the caller to finish, say 'let me look that up,' and only then call a tool," the company wrote. Phonic-1 can act mid-sentence, "so the answer is ready when they stop."
The model is trained on more than 50 languages and can switch languages mid-call while keeping the same voice, Phonic says.
How it compares
Third-party leaderboards back Phonic's ranking claims, though the margins vary.
| Benchmark | Phonic-1 | Grok Voice 3.0 | Note |
|---|---|---|---|
| Artificial Analysis Speech to Speech Index | 87.1 (No. 1) | 86.9 | Very narrow lead |
| Artificial Analysis Tau-Voice (agentic) | 79.9% | 71.6% | Clear lead |
| Cekura live phone calls (pass on all 3 runs) | 90.2% | 81.7% | Grok replied faster: 1.34s vs. 1.63s median |
| Cekura cost per minute | $0.060 | $0.080 | At published rates |
On Sierra's multilingual full-duplex τ³-Voice benchmark, Phonic reports 83.2, ahead of OpenAI's GPT-Live-1 at 79.9 and Google's Gemini 3.8 Live at 68.5.
Cekura's agentic phone tests cover clinic bookings and Medicare inquiries. Phonic-1 ranked first overall, while Grok Voice 3.0 was faster and handled callers changing their minds better.
Phonic is also candid about one gap. In blind third-party listening tests, judges preferred Phonic-1 over rival models most of the time, but real humans still beat it about 60% of the time.
Pricing
Phonic's pricing page lists a Starter plan at $0.07 a minute with no minimum, and a Pro plan at $0.065 a minute with a $200 monthly minimum. Enterprise pricing is custom. The company says there are no LLM pass-through or per-token charges.
Customers and funding
Phonic says it powers calls on six continents for companies including Ashby, Maven AGI, Rho and an unnamed Fortune 50 IT infrastructure company. Its customers work in support, healthcare, recruiting and logistics.
"The speech-to-speech model was ready for production, and its voices are some of the most natural we've heard," said Abhik Pramanik, co-founder and VP of engineering at Ashby. He said Phonic powers Ashby's AI interviews in 25 countries.
First Round led the Series A, joined by Lux Capital, Felicis, Definition and Essence VC, bringing total funding to $31.5 million. Angel investors include Erik Bernhardsson of Modal and Clement Delangue of Hugging Face. Phonic describes itself as a small team of researchers and engineers from MIT, Stanford and CMU, founded more than two years ago.
Why it matters
A small startup now tops voice benchmarks where OpenAI, Google and xAI compete. For developers building phone agents, tool calls during the caller's turn could cut dead air. Per-minute pricing also makes costs easier to predict than token billing.
What to do:
- Test on your own calls. Leaderboard leads are narrow on some measures.
- Compare latency and accuracy against Grok Voice 3.0 if speed matters most.
- Check total cost using Pro pricing if you expect sustained volume.
More on voice AI is in our Artificial Intelligence hub.
How we reported this
We checked Phonic's claims against its announcement, its pricing page and the Artificial Analysis and Cekura leaderboards. Sierra's τ³-Voice figures are as reported by Phonic. We cut the "10-person" headcount, which appeared only in an investor's social post, and demo and listening-test figures we couldn't find in fetchable primary text.
Published and fact-checked October 29, 2026.
Sources
- Phonic: Introducing Phonic-1 and our $27.5M Series A, Oct 29, 2026.
- Phonic: Pricing, accessed Oct 29, 2026.
- Artificial Analysis: Speech to Speech leaderboard, accessed Oct 29, 2026.
- Cekura: Speech-to-speech model benchmarks, updated Oct 22, 2026, accessed Oct 29, 2026.
DailyMech used AI tools to help research and draft this article; it was fact-checked by our editors.