The new voice of GPT chat: this is how the conversation will sound with the new generation

The new system is based on two models called GPT-Live-1 and GPT-Live-1 mini. The full version is intended for subscribers to the paid plans, including Go, Plus and Pro, while the smaller version is expected to be the default for users of the free plan. The rollout has begun worldwide on the Giphy Chat website and apps for iPhone and Android devices, but availability may gradually open up depending on the country and app version.

Two-way communication

Unlike the old voice mode, the new system does not have to wait until the user finishes a sentence and only then start processing it. The models work with a technology known as full two-way communication, where the system listens and speaks at the same time. This way, she can decide several times per second whether to continue listening, stop, talk, wait or activate another tool.

The meaning for the user may be felt from the first sentence. You can start talking, stop in the middle to think, return to a previous topic or interrupt the system while it is answering. Instead of interpreting every moment of silence as the end of the question and bursting into the user’s words, the system should understand better if it is a natural pause, a hesitation or a real end of the sentence.

This is a profound change from the first editions of the voice assistants. In the past, the user’s words were first transmitted to a system that converted the voice into text. A language model then formulated the answer, and a third system read it out loud. This chain caused delays, the loss of some of the tone and emotion in the speech and sometimes also answers that sounded mechanical.

The new models should also deal better with street noise, nearby conversations and changes in the rate of speech. OpenAI claims that the system can focus on the user’s voice instead of reacting to every sound picked up by the microphone. In internal tests conducted by the company, users preferred the new models over the previous voice mode in conversations that lasted between 5 and 10 minutes.

Not just small talk

The new voice is not just meant to make small talk. When a question is asked that requires a web search, deep thinking or performing a complex action, the voice system can transfer the task to the advanced GPT-5.5 model. While the other model searches and analyzes the information, the voice conversation can continue without getting stuck on hold.

The users of the appropriate programs will be able to choose between an immediate answer and medium and high thinking levels. In the immediate mode, the system will prefer to respond quickly. At the higher levels it will pass the request to a thinking model, which will spend more time analyzing the problem, but will still be able to return the result into the same voice conversation.

The system can also display visual information at the same time. During a conversation about the weather, stocks, sports or nearby places, the chat may display cards, forecasts, maps or data on the screen while the answer is spoken aloud. It is also possible to combine text messages and pictures in the same conversation, without opening a new conversation.

One of the outstanding capabilities of the new generation is live translation. The user can ask the assistant to translate a conversation between two people speaking different languages, while the system listens and transmits the things in real time. However, OpenAI admits that performance is still uneven across languages. Some may have a foreign accent, and the wording may be formal or less natural than the local speech.

what next

The company has also developed response rules adapted to teenagers and mechanisms for situations where the conversation turns to topics of self-harm, emotional dependence or mental crisis. In appropriate cases, the system may display written help resources and offer expert-tested support.

Despite the upgrade, some abilities will remain in the first stage outside of the new mode. Screen sharing and live video streaming are not supported at launch, and users who need them will be able to temporarily switch to the previous advanced voice mode. The new mode is not available at launch in business, enterprise, and educational environments, nor in ad hoc conversations, personalized chats, and desktop applications.