Before the Call Is Over, Phonic's Model Is Already Looking Things Up
Phonic launched its Phonic-1 voice model, which can call tools and look things up while the other person is still speaking, alongside a US$27.5 million Series A led by First Round. The model supports more than 50 languages.
- 01THE OLD STACK — The traditional voice stack runs in series, and pauses are unavoidable.
- 02THE CHANGE — Phonic-1 calls tools in parallel during the call.
- 03THE COVERAGE — More than 50 languages supported, switchable mid-call.
WHAT HAPPENEDWhat Happened
Phonic-1's core selling point is calling tools while a call is in progress. A traditional phone voice stack runs in series: listen first, then understand, then look things up, then speak, with heavy pauses and hand-off costs in between. Phonic moves tool calls into the live call, so the model can have results ready before the other person has finished speaking.
Language coverage is the second selling point. The company says the model supports more than 50 languages and can switch languages within a single call. For multinational customer-service and booking scenarios, switching language without redialling is a real cost saving.
The funding structure is equally clear. The US$27.5 million Series A was led by First Round, with Lux Capital, Felicis, Definition and Essence VC participating. Including earlier rounds, the company has raised about US$31.5 million in total, with customers across six continents.
THE PAUSE PROBLEMThe Real Bottleneck for Voice Assistants Is the Wait
The hardest part of a phone call to tolerate is silence. If the model only starts looking things up after listening, silence is unavoidable. Calling tools during the call runs lookup and listening in parallel, compressing the silence to a minimum.
That design puts heavy demands on the tools. When a tool returns an error, times out or responds in the wrong format, the model must handle it on the spot, mid-call, rather than retrying at leisure as in a text chat. Reliability matters more than cleverness here.
HUMANS STILL WIN 60 PERCENTHumans Still Win 60 Percent: The Gap Beyond the Leaderboards
The company's own evaluation concedes that human voices still win about 60% of preference comparisons. Speed and task completion can be quantified; tone, empathy and on-the-spot judgment cannot, and preference comparisons measure exactly the latter.
Third-party leaderboards point the same way. Phonic 1 tops the Artificial Analysis voice leaderboard at 87.1 points, with the fastest first audio at 0.61 seconds, but human preference is a different race. The two leaderboards should be read separately.
FROM MODEL TO SYSTEMThe Voice Race Is Shifting from Models to Systems
In the same week, Rime launched Prism, Synthflow closed a US$20 million Series A, and Scaled Cognition released an enterprise voice model. Voice models are no longer single-point breakthroughs; the competition is over the whole phone system.
The deciding factors are in the deployment details: noise, accents, interruptions, packet loss, tool timeouts, data retention and launch commitments. None of these appear in leaderboard scores, yet they decide whether customers renew.
THE SHIFTWhat Changed
SIGNAL JUDGMENT — ANALYSISVoice Assistants Compete on Waiting Less, Not Talking Better
Tokenta Signal's judgment is unchanged: Phonic-1's value lies in the waiting time saved through parallel tool calls. The real test is how it behaves on the phone when a tool fails, not its leaderboard score.
WHAT REMAINS UNCERTAINWhat Remains Uncertain
EVIDENCEEvidence
The original report confirms Phonic launched the Phonic-1 voice model, which can call tools during a call and supports more than 50 languages, alongside a US$27.5 million Series A led by First Round.
PRIMARY SOURCESPrimary Sources
- Precise Impact: AI Daily Brief, 31 October 2026
https://www.preciseimpact.ai/briefing/ai-news-digest-2026-10-31 - AlphaSignal: MIT-Founded Phonic Beats Grok and GPT With Fastest Voice AI Response
https://alphasignal.ai/news/mit-founded-phonic-beats-grok-and-gpt-with-fastest-voice-ai-response
RELATEDRelated Intelligence
- Microsoft Winds Down Its Oracle AI Cloud Rental as Older Nvidia Chips Fall Behind
- An Unreleased Anthropic Model Proved an 18-Year-Old Quantum Conjecture, Machine-Checked in Lean
- IBM Unveils Quantum System Two in Chicago, Its Most Advanced Computer Yet, Powered by Nighthawk
- Vistra Posts $1.14B Q3 Profit as Power EBITDA Jumps Over 50%, Lifts 2026 Outlook