PhoneLLM makes voice agents a $0.0025-minute bet

Pipecat released PhoneLLM Alpha 1, an open-weights voice-agent model with 3.5B active parameters that its release announcement says matches GPT-5.6 Terra at 94% lower cost and 1,300ms faster P95 time-to-first-token. The BSD-licensed model runs at a reported $0.0025 per agent-minute with 88 concurrent agents on one B200; that is a vendor benchmark, not a production guarantee. Teams building phone support should test it against the 1,500ms voice budget and tool-call accuracy—not generic chat scores—alongside the archive’s cost analysis of grounded ChatGPT voice.