June 6, 2026. OpenAI expanded its Realtime API with GPT-Realtime-2, the first voice model with GPT-5-class reasoning. In plain terms, AI voice agents can now listen, reason, respond, translate, and take action while a call is still happening. The "AI receptionist" finally works well enough to put on a real phone line.
What OpenAI shipped
- GPT-Realtime-2. GPT-5-class reasoning in live voice, with the context window expanded from 32K to 128K tokens, so it handles harder, longer conversations naturally.
- GPT-Realtime-Translate. Live translation from more than 70 input languages into 13 output languages, keeping pace with the speaker.
- GPT-Realtime-Whisper. Streaming speech-to-text that transcribes a conversation live as the speaker talks.
- Real deployments already. Early testers include Zillow, which is building an assistant that reasons through complex voice property searches, plus Priceline and Deutsche Telekom.
What it means for operators
The bottleneck on voice agents was never the idea. It was latency and reasoning, agents that interrupted, misheard, or could not handle anything off-script. GPT-5-class reasoning in real time fixes the off-script problem, and live translation opens multilingual markets, which is a big deal for a business serving English, Arabic, and Hindi callers from one line. Missed calls are the most expensive leak in service businesses, and this finally closes it.
How to build with it
- Put a voice agent on your inbound line for after-hours answering, lead qualification, and appointment booking.
- Use the translation model to serve multilingual customers with a single agent.
- Integrate it with your calendar and CRM so the agent books and logs every call automatically.
This is exactly what we build as AI phone agents inside AI automation and through GoHighLevel voice agents.
The bottom line
Voice was the last place AI felt clumsy. That just changed. The businesses that put a reliable agent on their phone line this year will stop losing the leads that used to go to voicemail.
Frequently Asked Questions
It is OpenAI’s new voice model that brings GPT-5-class reasoning to live conversations, letting agents listen, reason, respond, and act in real time, with the context window expanded to 128K tokens.
For bounded use cases, yes. Appointment booking, lead qualification, after-hours answering, and FAQs work reliably now. Open-ended sales still benefits from human handoff, so the right pattern is agent-first with escalation.
Yes. OpenAI’s GPT-Realtime-Translate handles live translation from more than 70 input languages into 13 output languages, so one agent can serve callers in several languages.
Answer inbound calls around the clock, qualify leads, book appointments straight into your calendar, and follow up, so you stop losing leads to voicemail and after-hours gaps.
Through integrations. We wire the agent into your CRM and calendar so every call is logged and every booking is created automatically, with no manual data entry.
No. A partner can configure, integrate, and launch the agent for you in a few weeks, then tune it against real calls.