Summary
On July 6, 2026, OpenAI released two updated Realtime API voice models, gpt-realtime-2.1 and gpt-realtime-2.1-mini, cutting p95 latency by at least 25% and improving speech recognition, noise handling, reasoning, tool use, and instruction following for low-latency voice and multimodal agents.
What changed
OpenAI released gpt-realtime-2.1 and gpt-realtime-2.1-mini for the Realtime API, delivering at least 25% lower p95 latency across Realtime voice models plus better recognition, noise handling, reasoning, tool use, and instruction following.
Why it matters
Latency is the dominant UX constraint for voice agents; shaving at least 25% off p95 makes real-time, tool-using voice assistants and phone agents feel more natural and lets developers push more reasoning into a turn without breaking the conversational loop. A cheaper mini tier widens the range of production voice workloads that are economically viable.
Evidence excerpt
OpenAI released gpt-realtime-2.1 and gpt-realtime-2.1-mini on 2026-07-06 with at least 25% lower p95 latency across Realtime voice models, plus improved recognition and noise handling.