OpenAI Connects GPT-Live to Agents on Mobile and Web

OpenAI connects its voice model GPT‑Live to models capable of performing tasks in applications and browsers, now available from the ChatGPT app and website. The company describes a usage that alternates between voice and text depending on the context, emphasizes mobility, and prepares for a broader meeting around personal agents.

Voice Input, Text or Audio Output Depending on Context

The team in charge of the voice mode indicates that they have taken into account the fact that speaking is often faster than typing, but reading remains faster than listening. Therefore, they adapt ChatGPT's responses to audio requests, sometimes choosing text, sometimes voice, or a mix of both. Another challenge identified concerns adoption in the workplace, where workstations remain centered on computers, while connected speakers persist.

Delegating Tasks by Voice, Including on the Go

The new features allow users to verbally delegate unique or recurring tasks, even complex ones involving multiple steps, a browser, and different applications, all while on the move. During a demonstration, Atty Eleti, product lead for ChatGPT's voice, used only his voice to analyze his expenses on DoorDash, book a pickleball court for $5, and reorganize his calendar. According to him, the main advantages of voice mode are speed, the possibility of hands-free use, and assistance with tasks like language learning or speech preparation, activities he practices during commutes or walks.

GPT‑Live Connected to Agentic Models on Web and Mobile

OpenAI is deploying its most advanced voice system on the ChatGPT website and mobile app. This version connects GPT‑Live to artificial intelligence models capable of performing tasks on computers and in browsers, continuing a trajectory already initiated before the rise of personal agents. GPT‑Live, capable of listening and responding simultaneously, was previously accessible for ChatGPT's standard conversational mode and connected to agentic models via the desktop application. In July, Greg Brockman unveiled an audio model that also offered this simultaneous dialogue feature.

Popularity, Competition, and Hardware Horizon in 2027

According to Atty Eleti, over 150 million users are already utilizing the dictation and voice mode features of ChatGPT. OpenAI is expanding its offerings as personal assistants like Meta's Instinct and Muse see their usage grow, although they do not yet provide a direct alternative to these specialized solutions. The company is ramping up announcements ahead of OpenAI DevDay scheduled for next week, and it is reported that a smart speaker is in preparation for 2027. OpenAI also expresses the ambition that voice will ultimately become the primary mode of interaction with AI.