Phonic-1 Wants Voice Agents to Act While You're Still Talking, and It Just Raised $27.5M to Prove It
Primary source: Phonic's official blog post "Introducing Phonic-1 and our $27.5M Series A" (published Oct 29, 2026; checked Oct 30, 2026)
Oct 30, 2026: Most AI phone agents share one tell: you finish talking, there's a pause, and the bot says "let me look that up." Voice startup Phonic says its new model removes that pause. On Oct 29, 2026, the company launched Phonic-1, a speech-to-speech model built for agentic tasks, and announced a $27.5M Series A at the same time.
Phonic's headline claim is that Phonic-1 is "the only model that can act while the caller is still talking, so the answer is ready when they stop." That is the company's own claim, and we have not tested it independently.
What Phonic-1 is
Phonic-1 is a speech-to-speech model: audio goes in and audio comes out. Phonic contrasts this with the usual "cascade" approach, where one model transcribes the caller, a second decides what to say, and a third speaks the reply.
Phonic argues a transcript "drops the frustration in a caller's voice, the long pause before an answer, even the way someone says their own name." The company says it has worked only on speech-to-speech since its founding more than two years ago.
The four big claims (all Phonic-reported)
- Tool calls mid-sentence. Phonic says other models wait for the caller to finish before calling a tool, while Phonic-1 can start acting while the caller is still speaking.
- Speed plus reasoning. Phonic says Phonic-1 took the number one spot on the Artificial Analysis Speech to Speech Index and "stands alone on the Pareto frontier of intelligence and latency" on end-to-end full-duplex benchmarks, citing Artificial Analysis (Oct 29, 2026).
- Naturalness. Phonic says that in blind human evaluations run by a third party, judges preferred Phonic-1 "the majority of the time." The company also admits that against real humans, "humans win ~60% of the time."
- Multilingual. Phonic says the model is trained on 50+ languages and can switch languages mid-conversation while keeping the same voice.
The benchmark numbers
The only hard benchmark figure in the launch post is for Sierra's multilingual full-duplex τ³-Voice benchmark. These scores are as reported by Phonic:
| Model | τ³-Voice score (Phonic-reported) |
|---|---|
| Phonic-1 | 83.2 |
| GPT-Live-1 | 79.9 |
| Gemini 3.8 Live | 68.5 |
Phonic cites Artificial Analysis, Sierra τ³-Voice and Cekura for its charts. We have not checked those leaderboards directly, so treat these rankings as company claims.
A 44-minute demo call
To show off long, tool-heavy calls, Phonic published a demo of a 44-minute call with 123 tool calls. In it, a maid of honor phones a hotel in Brazil to book a bachelorette party and Phonic-1 handles the call.
The money and the customers
- Series A: $27.5M, led by First Round, with participation from Lux Capital, Felicis, Definition and Essence VC.
- Total raised to date: $31.5M, per Phonic.
- Angels named by Phonic include Clement Delangue (Hugging Face), Amjad Masad (Replit) and Guillermo Rauch (Vercel), among others.
- Customers named: Ashby, Maven AGI, Rho and "a Fortune 50 IT infrastructure company." Phonic says it handles calls on six continents in customer support, healthcare, recruiting and logistics.
Ashby co-founder Abhik Pramanik is quoted in the post saying Phonic "powers our AI interviews in 25 countries."
Why it matters
Phonic is betting on speech-to-speech for production calls, not just impressive demos. If "act mid-sentence" holds up in real deployments, it fixes the most noticeable weakness of today's phone bots: dead air while they look something up.
What's confirmed / not confirmed
Confirmed (Phonic's official post, Oct 29, 2026): - Phonic-1 launched Oct 29, 2026, as a speech-to-speech model for agentic tasks. - Phonic raised a $27.5M Series A led by First Round, bringing total funding to $31.5M. - Phonic says the model is trained on 50+ languages. - The published τ³-Voice scores: Phonic-1 83.2, GPT-Live-1 79.9, Gemini 3.8 Live 68.5.
Not confirmed / company claims only: - The "only model that can act while the caller is still talking" claim and the #1 Artificial Analysis ranking come from Phonic. We did not independently verify them on Artificial Analysis. - The blind human preference results come from an unnamed third party cited by Phonic. Full methodology is not in the post. - Pricing and public API availability details were not stated in the launch post we reviewed.
Related
Up: AI
