OpenAI announced a major upgrade to ChatGPT’s voice capabilities at a press briefing on Tuesday, introducing GPT‑Live‑1, the company’s “smartest voice model” to date. Unlike the previous turn‑based system, GPT‑Live‑1 operates in full‑duplex mode, meaning it can process spoken input and generate spoken output at the same time. The result, according to research lead Kundan Kumar, is a conversation that feels more like talking to another person and less like waiting for a machine to finish before you can speak.
The new model is designed to interrupt users less often. If you pause mid‑sentence, GPT‑Live‑1 waits for you to continue rather than cutting you off. When a question requires deeper reasoning or a web search, the model automatically routes the query to OpenAI’s most advanced text engines, such as GPT‑5.5, before delivering a spoken answer. This seamless hand‑off shortens the lag between asking and hearing a response.
Beyond smoother dialogue, GPT‑Live‑1 adds visual context to voice interactions. Users asking about the weather will see an AI‑generated forecast graphic; queries about stocks or sports scores trigger real‑time charts and scoreboards. Product lead Atty Eleti described the feature as “supplementing conversations with AI‑generated visuals,” a move that could make voice assistants more useful for quick data checks.
Real‑time translation is another headline feature. Previously, users had to finish speaking before the system could translate their words. Now the model translates on the fly, allowing bilingual conversations without awkward pauses. Users can also command the voice to stay silent until prompted, a capability that was not possible with the older model. When listening, GPT‑Live‑1 acknowledges with brief utterances like “mhmm,” “yeah,” or “got it,” signaling that it is still engaged.
OpenAI said the upgrade includes enhanced safety layers. Built‑in safeguards steer the model away from harmful responses and can end chats altogether in higher‑risk situations. The company highlighted two new support functions: expert‑vetted crisis‑helpline assistance for self‑harm discussions and age‑appropriate replies for teenage users. These measures come as OpenAI faces a series of lawsuits alleging that ChatGPT contributed to users’ mental‑health issues.
GPT‑Live‑1 is being rolled out across iOS, Android and the web. It will power the voice experience for ChatGPT Go, Plus and Pro subscribers, while a smaller, more efficient “mini” variant will serve free users by default. OpenAI expects the full‑duplex model to open the door for future features such as continuous conversation, richer multimodal outputs and tighter integration with the company’s expanding suite of AI tools.
Industry observers see the launch as a significant step toward making voice assistants feel less robotic and more conversational. By reducing interruptions, adding visual aids and enabling on‑the‑spot translation, OpenAI is addressing long‑standing complaints about clunky voice interactions. The company’s emphasis on safety and crisis support also signals a growing awareness of the ethical stakes surrounding conversational AI.
While the new capabilities are likely to attract more users to paid tiers, the rollout will be watched closely by regulators and consumer‑rights groups, especially given the pending litigation. OpenAI’s ability to balance innovation with responsible deployment will shape how quickly voice AI becomes a staple in everyday digital life.
Cet article a été rédigé avec l'assistance de l'IA.
News Factory APP - actualités agentiques pour booster votre SEO et AEO.