OpenAI has rolled out full duplex voice control for its Codex and ChatGPT desktop applications, letting users speak and listen at the same time. The update arrives as a prediction market gives the company just a 9% chance of having the best AI model by December 2026.
How the voice feature works
Full duplex means both sides can talk and hear simultaneously, like a real conversation. Earlier voice interfaces often used half-duplex, where one person speaks and then waits for a reply. With this change, users can interrupt, ask follow-ups, or talk over the AI without awkward pauses. The feature is built into the desktop versions of Codex, OpenAI's coding assistant, and ChatGPT, its general-purpose chatbot.
The 9% prediction
One prediction market now pegs OpenAI's odds of leading the AI model race by the end of 2026 at 9%. That's a low number for a company that kicked off the current generative AI boom with GPT-3.5 and GPT-4. The market's bet suggests traders see a crowded field of competitors closing the gap. OpenAI faces pressure from other labs, open-source projects, and well-funded startups, all racing to build the next breakthrough.
The 9% figure doesn't mean OpenAI will fall behind — it's a snapshot of what bettors think today. Prediction markets aggregate many individual guesses, and they can shift quickly as new models launch or benchmarks change. Still, the number stands out for a firm that has often been seen as the frontrunner.
OpenAI hasn't said whether full duplex voice will come to its mobile apps or web interface. The desktop-only launch suggests the company is testing the feature with a smaller audience first. Developers using Codex may find the hands-free interaction useful for dictating code or debugging aloud. ChatGPT users get a more natural way to brainstorm or ask questions without typing.
The company also hasn't disclosed how many people have tried the new voice mode since it went live. Adoption rates will likely influence whether OpenAI expands the feature to other products. For now, the update gives desktop users a taste of what real-time voice AI feels like — and a reminder that the race to build the best model is far from over.
