Phonic-1

Phonic-1 is a speech-to-speech model from Phonic, introduced on October 29, 2026 alongside a $27.5M Series A led by First Round. It is positioned for agentic voice agents: hear caller audio directly, speak naturally, and call tools while the caller is still talking, without a cascade of separate ASR → LLM → TTS stages.

Launch: phonic.ai/blog/phonic-1 · Docs: docs.phonic.co · Model id: phonic-1

Key points

1. Named model ship: Phonic-1 (API id phonic-1), not a generic "Phonic voice" rebrand.
2. Benchmark claim: #1 on Artificial Analysis Speech to Speech Index at 87.1 (High config), narrowly ahead of Grok Voice 3.0; strong on agentic task success with sub-600ms speech-in→speech-out latency.
3. Differentiation: mid-utterance tool calls so answers are ready when the caller stops; 50+ languages with mid-call language switch; SOC2 / HIPAA options.
4. Pricing: about $0.07 per minute via Phonic API / SDKs / CLI / agent builder.
5. Traction this window: X news + independent write-ups (OrcaRouter, Digg, LLMs Blog) the same day as the Series A post.

Why it matters

Clear proper noun with same-day official blog, API surface, and independent leaderboard placement—easy to track against GPT-Live / Gemini Live class voice stacks.

How to try it

1. Read the launch post and create an API key in the Phonic console.
2. Build an agent with model id phonic-1 via API, TypeScript/Python SDKs, CLI, or the no-code builder.
3. Compare agentic call success and latency on your own telephony traffic against cascade stacks.

Confirm region availability and enterprise BAAs before production healthcare/finance traffic.