Introducing PhoneLLM: Open Source Voice Model with 1/3 Latency and 1/18 Cost of GPT-5.6 Terra

charles_irl · x · 2026-08-28

Introduces PhoneLLM, an open model designed for voice agents. It achieves GPT-5.6 Terra performance on typical voice tasks but with 1/3 of the latency and 1/18 of the cost.

Voice agents require models that are both low-latency and capable at tool calling and instruction following. PhoneLLM addresses this trade-off, enabling voice-speed responses without relying on "thinking" models.

Original post →

More from Models

Models channel →