OpenAI is releasing two new voice fashions at this time referred to as GPT-Stay-1 and GPT-Stay-1 mini the corporate says carry main enhancements to ChatGPT’s voice performance. You might recall the corporate launched Superior Voice Mode within the summer of 2024. On launch, the function was spectacular and felt like a significant improve over legacy voice assistants like Siri and Alexa, however over time its limitations started to really feel extra pronounced. Particularly, Superior Voice Mode employed what OpenAI describes as a “turn-based” voice mannequin, which features precisely prefer it sounds. The mannequin has to attend for the consumer to cease talking earlier than it could possibly reply. As you may think, this typically resulted in a stilted back-and-forth. What’s extra, as a result of the mannequin used silence as its cue to start out speaking, it could typically mistakenly interpret a short pause as the top of a immediate or query, resulting in awkward interruptions.
Against this, the GPT-Stay-1 household of fashions are constructed on high of a duplex structure. They will concurrently course of inputs whereas producing an output on the identical time. “This enables the mannequin to interact in a extra pure back-and-forth, keep a greater sense of time, and even carry out dwell translations,” explains OpenAI. On the identical time, the corporate has designed GPT-Stay to delegate duties. If the brand new voice fashions decide your query or immediate would profit from reasoning, an online search or different, extra agentic capabilities, they will flip to OpenAI’s different fashions, together with GPT-5.5, for help. These methods will work within the background, permitting you to proceed your dialog with GPT-Stay.
Collectively, these modifications lead to a voice expertise OpenAI believes folks will discover extra helpful. You’ll be able to interrupt the brand new voice fashions at anytime, and even ask them to decelerate if you happen to discover they’re speaking too quick. As you are talking, they may also verify they’re listening with verbal acknowledgements like “mhmm” and “bought it,” and OpenAI says it has made the brand new fashions higher at focusing when there are distracting noises within the background. ChatGPT Voice can now additionally generate visuals, within the type of widget-like playing cards, for subjects just like the climate, sports activities, shares and extra. As earlier than, assist for picture and file uploads is constructed into the function, and it’s also possible to use it to look the online.
On the identical time, OpenAI says it has constructed new safeguards into ChatGPT Voice. “When the system detects probably unsafe output, it could possibly steer the mannequin towards a safer response, floor extra security messaging or finish the voice dialog in higher-risk instances,” the corporate says. Dad and mom and guardians can use OpenAI’s just lately launched parental controls to disable ChatGPT Voice for his or her teenagers. In the event that they’ve given their youngster entry to ChatGPT Voice and the system detects they’re making an attempt to push the dialog towards self-harm, the system will notify the mother or father.
OpenAI is beginning to roll out GPT-Stay to ChatGPT on Android, iOS and the online, starting at this time. GPT-Stay 1 would be the go-to mannequin for these paying for Go, Plus and Professional subscriptions, whereas Free accounts will default to the smaller Stay-1 mini mannequin.
