OpenAI ships GPT Realtime 2.1 for lower latency voice agents

The new flagship and a cheaper mini variant cut p95 latency by at least 25 percent with stronger reasoning, tool use, and instruction following, aimed squarely at production voice agent workflows.