← Back to Voice models
PhonicNotable
Phonic-1
Released Oct 29, 2026
Phonic's end-to-end speech-to-speech model for agentic voice tasks, built to call tools while the caller is still speaking. It's served on Phonic's voice-agent platform.
- Price per minute
- $0.070/min
- Inputs
- audio
- Tool calling
- Yes
- Languages
- 50
- Voices
- —
- Big Bench Audio
- 98.9%
- Access
- Generally available
- Licence
- Proprietary
Strengths & weaknesses
Strengths
- #1 on AA's Speech to Speech Index (87.1) with Big Bench Audio 98.9%
- Calls tools while the caller is still speaking
- 0.61s time to first audio (AA)
- Flat $0.07/min with no LLM or token pass-through
Weaknesses
- Platform-bound: no public model ID, and Enterprise is needed for self-hosting or VPC
- Phonic's τ³-Voice comparison is self-reported
Notable: top 10 on both AA boards (#1 S2S Index 87.1; #3 Big Bench Audio 98.9%, listed as 'Phonic 1 (High)'). Phonic reports τ³-Voice multilingual full-duplex 83.2 vs GPT-Live-1 79.9 and Gemini 3.8 Live 68.5. Languages: trained on '50+', stored as 50. The pricing page lists Starter $0.07/min, Pro $0.065/min ($200 monthly minimum) and custom Enterprise, without tying them to Phonic-1 by name. phonic.co now redirects to phonic.ai. The old docs at docs.phonic.co listed usage pricing 'starting at $0.15 per minute', which looks outdated.