OpenAI launches GPT-Live real-time speech model with simultaneous listening and speaking

OpenAI released GPT-Live on July 8, a real-time speech model (available in GPT-Live-1 and mini variants) that can listen, process, and respond while the user is still speaking — no need to wait for sentence completion. The model can interject with brief responses like 'mm-hmm' and 'got it' mid-conversation, and seamlessly hand off complex reasoning tasks to GPT-5.5 without the user noticing. GPT-Live mini adds reasoning and tool calling capabilities at no extra cost.