Phonic-1: speech-to-speech model that calls tools mid-call
Phonic-1 is a speech-to-speech model for voice agents that runs tool calls mid-conversation. Released October 29, 2026, with per-minute pricing.
On October 29, 2026, Phonic released Phonic-1, a speech-to-speech model for voice agents that call external tools. The company says the model ranks first on the Artificial Analysis Speech to Speech Index.
What’s new
Most voice agents chain separate speech recognition, language, and speech synthesis models. Phonic-1 processes audio end to end and can start tool calls while the caller is still talking. In the company’s travel-booking demo, it made five tool calls and began replying 1.1 seconds after the caller stopped, while GPT-Live-1 took 15.7 seconds on the same script. Phonic cites latency around 600 ms, support for 50+ languages, and mid-call language switching with the same voice.
How to try it
The model is available now, and sign-ups before November 30 receive $250 in free credits. Documentation lives at docs.phonic.ai. The homepage FAQ lists Starter at $0.07 per minute and Pro at $0.065 per minute with a $200 monthly minimum; Enterprise pricing is custom.
Limitations
The launch benchmarks come from the company, Artificial Analysis, and Sierra’s τ³-Voice. In blind listening tests run by Podonos, human voice actors beat Phonic-1 57% to 32%, and Phonic admits humans still win about 60% of naturalness comparisons. The launch post contains no prices, so the figures above come from the homepage.
Prepared with AI from public sources. Reviewed by the site’s author.