OpenAI's GPT-Live: A ChatGPT voice that both listens and speaks.
OpenAI encourages users to engage with ChatGPT through conversation rather than typing. On July 8, it introduced GPT-Live, a new generation of voice models that the company claims enhance the realism of interactions with AI. From today, ChatGPT users around the globe can access two versions: GPT-Live-1 and a smaller variant, GPT-Live-1 mini.
The main improvement lies in the architecture. GPT-Live operates in “full-duplex” mode, allowing it to listen and speak simultaneously. It can interject with phrases like “mhmm” or “got it,” facilitating quick exchanges while also being able to pause for your thoughts. Previous voice models were required to wait until a user finished speaking, which often resulted in interruptions at inopportune moments.
The second noteworthy advancement is delegation. When a query requires a web search or complex reasoning, GPT-Live refers it to a more powerful model in the background, specifically GPT-5.5 at launch. While it waits for the response, GPT-Live continues the conversation, seamlessly integrating the answer later.
“This reflects the way humans communicate,” said Atty Eleti, the ChatGPT voice product lead, during a media briefing reported by CNET. “We maintain the dialogue while we process in the background.”
One demonstration stood out, showcasing the model's capability of real-time, simultaneous translation, providing ongoing translations as a presenter spoke. TechRadar deemed this feature genuinely valuable. Additionally, the model responds to a wake word, with staff using “Hey Chat” in their demo.
OpenAI claims GPT-Live is its most intelligent voice model to date. Users can select a reasoning level—Instant, Medium, or High. The voice can now generate visual cards for updates on weather, stocks, and sports. The company asserts that GPT-Live-1 and the mini outperform the previous Advanced Voice Mode in test conversations lasting five to ten minutes and show significant improvement in benchmarks related to scientific reasoning and agentic web searches.
According to OpenAI, over 150 million individuals use ChatGPT’s Voice and Dictation features weekly. GPT-Live-1 will be the default for Go, Plus, and Pro subscribers, while free users will have access to the mini version. The rollout will be available on iOS, Android, and the web, although it may take a few days to become widely accessible.
The ability to sound human has its pros and cons. OpenAI reports that it has implemented voice-specific safety training and real-time safeguards that can guide responses, offer crisis resources, or terminate conversations in higher-risk situations, with additional protections for teenagers. The company also emphasizes that GPT-Live uses preset voices and does not aim to mimic any real individual.
This caution is significant; a model that resonates with users as a friend may build trust more easily but can also mislead. Critics, such as Meredith Whittaker from Signal, have cautioned against considering chatbots as genuine friends. The competition remains intense, with rivals like ElevenLabs and numerous voice-agent startups pursuing similar goals. This competitive environment has prompted Amazon to revamp Alexa.
This update coincides with OpenAI's launch of a widespread rollout of GPT-5.6, part of an accelerated release strategy. Historically, voice has been a supplementary feature rather than the primary method for users to interact with ChatGPT. With GPT-Live, OpenAI is wagering that conversation will become the preferred mode over typing. However, it remains to be seen whether users desire an AI that is always listening.
Other articles
The GhostApproval bug disrupts six leading AI coding agents.
Wiz's 'GhostApproval' employs an outdated symlink technique to enable six AI coding agents, ranging from Amazon Q to Cursor, to operate beyond the sandbox and transfer ownership of the box.
According to Gartner, customers favor ChatGPT over company chatbots.
According to Gartner, customers are three times more inclined to utilize third-party Generative AI tools like ChatGPT instead of company chatbots for support, as the returns on AI investments are delayed compared to the expenditures.
Apple begins testing more affordable Chinese RAM in iPhones, but your wallet won't benefit from the savings.
Apple is evaluating memory chips from CXMT, a Chinese chip manufacturer that is supported by the government and is on the Pentagon's watchlist. Washington has previously opposed this and is expected to do so again.
AI security cameras might soon identify your walking pattern before they are able to recognize your face.
A novel AI gait recognition system can recognize individuals based on their walking patterns, providing security cameras with an additional long-distance signal when faces are unclear, obscured, or too small to rely on.
Meta might be developing smart glasses that are capable of always seeing and hearing, equipped with advanced AI sensing technology.
Meta is said to be experimenting with “super sensing” AI glasses designed to gather comprehensive data about your life and provide it to an AI assistant that will aid you.
Only 6% of managers in the UK believe that Gen Z is prepared for the workforce.
A CMI report reveals that 45% of UK Gen Z believe they are prepared for the workforce, while only 6% of managers concur. Furthermore, 82% of these managers were promoted without any training.
OpenAI's GPT-Live: A ChatGPT voice that both listens and speaks.
OpenAI has introduced GPT-Live, a voice feature for ChatGPT that allows for simultaneous listening and speaking, complete with live translation. This feature is being made available to all users, including those using the free version.