OpenAI is now proudly launching two new models, GPT-Live-1 and GPT-Live-1 small versions, that completely reinvent ChatGPT’s voice assistant features.

As you may remember, the company introduced the Advanced Voice Mode for the first time in the summer of 2024. This feature, which initially felt like a major revolution compared to classical assistants such as Apple’s Siri or Amazon’s Alexa, became the subject of criticism over time due to some of its limitations.

Especially the fact that the old generation system waited for you to finish your speech before responding and intervened by thinking that small pauses meant “the sentence was finished” significantly disrupted the naturalness of the conversation.

ChatGPT Changes the Rules of Chat: What Does Duplex Architecture Offer?

Voice models used in the past, called “turn-based”, literally had to wait for their turn. The assistant had no chance to take action or intervene until you finished speaking.

However, the new GPT-Live-1 family uses a concurrent process architecture called duplex technology. Thanks to this innovation, the model simultaneously analyzes what you say and creates its own answer within seconds.

According to the statement made by OpenAI officials, this system allows for a fluent mutual conversation, while also successfully handling difficult tasks such as simultaneous translation.

Now, while talking with ChatGPT, you can interrupt the assistant whenever you want. More interestingly, if you think that the assistant is talking too fast or that you cannot follow a technical subject he is explaining, you can warn him and ask him to talk slower.

While you are explaining something, the system makes you feel that it is actively listening to you by making human-like verbal affirmations such as “uh-huh”, “I understand” and “yes”. In addition, the company emphasizes that the new version does a much better job of filtering out disturbing background noise before.

Task Deployment and Background GPT-5.5 Support

One of the most striking features of the new sound model is its ability to delegate tasks independently. If GPT-Live-1 detects that your question requires deeper reasoning, an extensive internet search, or a different analytical approach, it quickly transfers the mission to other models from OpenAI.

Especially at this point, the system gets help from the GPT-5.5 model, which offers huge power in the background. While you continue your conversation fluently with the assistant, these background systems collect the necessary information and seamlessly integrate it into the assistant’s conversation.

Visual Cards and Advanced Security Measures

Not only limited to voice communication, the new ChatGPT Voice can now create widget-like interactive visual cards directly on your screen for bets such as weather forecast, sports match scores, stock market data.

Just as we were used to in previous versions, the feature of uploading photos and documents to the system continues to work uninterruptedly via the same infrastructure.

On the security side, OpenAI is taking new multi-layered measures to protect families and young users. When the system detects potentially risky or harmful content, it directs the model to a more confident answer.

In much higher risk situations, it has the authority to terminate the voice chat directly. Thanks to the parental controls that the company recently introduced, families can completely disable the ChatGPT Voice feature in their children’s accounts.

If the assistant detects that young users are engaging in dangerous conversations that could harm themselves, the system immediately sends a warning notification to parents.

With this huge update, which is now distributed globally on Android, iOS and web platforms; Go, Plus and Pro subscribers receive the GPT-Live-1 model directly. Free version users step into this impressive audio artificial intelligence experience with Live-1 small.