The pipecat-ai project has released the phonellm-alpha-1 model, which achieves performance comparable to GPT 5.6 on typical voice agent tasks.
- The model operates at one-third the latency of the comparison baseline.
- It costs approximately one-eighteenth of the reference solution.