PhoneLLM Max Alpha 2

Pipecat’s larger voice-agent model, with reasoning, its 1M-token context window, and tool calling for multi-turn customer conversations.

Talk to an engineer

Model details

Developed by
Pipecat
Model family
PhoneLLM
Input modalities
Text
Output modalities
Text
License
OpenMDW-1.1

View repository

PhoneLLM Max Alpha 2 is a larger language model from Pipecat, built on the text-only backbone of NVIDIA's Nemotron 3.5 Super VL model. With approximately 120 billion total parameters and a 1M-token context window, it combines multi-turn conversation and tool-calling capabilities with thinking enabled by default. Max generates reasoning before its final response, supporting voice-agent workflows that call for a short reasoning step before each reply or tool call. Compared with Mini, it requires more serving resources and introduces additional latency before the visible answer.