OpenAI launches three new real-time audio API models
…The model supports a 128K token context window, up from 32K in the previous version, which makes longer voice sessions and complex multi-step agentic flows practical without external memory scaffolding. What…
