OpenAI is close to releasing a major upgrade to ChatGPT's voice mode. The new model, currently tagged Bidi 1, is built to listen and speak at the same time, rather than stopping every time the user makes a sound. No official announcement has been made. The model surfaced through code references and UI sightings inside the ChatGPT app, with a small group of users already testing it.
The current voice mode runs on a turn-based system. When you say anything mid-response, even a casual "okay," it stops the AI completely and restarts the cycle. That is the core problem Bidi 1 solves. It absorbs interruptions, issues brief acknowledgments like a human would, and keeps going without losing its place.
The second problem it fixes is memory. The existing voice mode tends to drop earlier parts of a conversation as it runs on. Bidi 1 holds the full thread. That alone makes it meaningfully more useful for anything beyond a quick one-off question.
The new mode will sit alongside the current Advanced Voice Mode as a separate option, not replace it. Users will be able to toggle between them. It also introduces three response tiers: High, Medium, and Instant, mirroring what already exists on the text side of ChatGPT, so users can trade reasoning depth for speed depending on the task.
Real-time translation is also expected to arrive with Bidi 1, which opens up a specific set of uses: customer calls across language barriers, multilingual team briefings, and similar scenarios that currently require either a human interpreter or a clunky back-and-forth.
Why does this matter beyond ChatGPT itself? The voice AI market is moving fast. Production deployments of AI voice agents grew 340% year-over-year across more than 500 organisations, according to industry tracking data. Eighty percent of businesses plan to integrate voice AI into customer service by the end of 2026. The bottleneck has not been willingness; it has been quality. Voice tools that interrupt awkwardly, lose context, or struggle with natural conversation create more frustration than they solve.
Bidi 1 does not ship as an enterprise product. It is a consumer ChatGPT feature first. But OpenAI has signalled that API access may follow, which is when it becomes a building block that businesses can embed in their own products: phone support lines, internal help desks, field tools for sales or claims teams, and so on.
The practical read for now: this is not something to act on immediately. Wait for the full release and the API timeline before evaluating it against your own needs. What it does signal is that the gap between a good voice conversation and a good AI voice conversation is closing faster than most people expected a year ago. Anyone with a contact centre, a customer phone line, or a staff-facing tool that relies on voice should start forming a view on where that fits in their operations, because the technology will be there before the internal plan is.