OpenAI to retire legacy voice modes in favor of GPT-Live
ChatGPT will phase out Advanced and Standard voice options by December 2026, moving all users to a new system that allows for simultaneous listening and speaking.
The short version
- Advanced and Standard voice options will be removed from the model picker on December 11, 2026.
- GPT-Live allows for simultaneous listening and speaking, removing the need for turn-taking.
- The new system can delegate complex tasks to text models while maintaining an active voice conversation.
- Live video and screen-sharing features will continue to use the old Advanced voice infrastructure for now.
Why it matters
This shift moves AI interaction away from 'walkie-talkie' style turn-taking toward a more human-like flow. It also allows the AI to perform heavy lifting in the background without breaking the conversational loop.
The end of legacy voice
OpenAI has announced it will discontinue two of its primary voice options in ChatGPT. Both Advanced and Standard voice will be removed from the model picker on December 11, 2026. This move marks a transition toward a single, unified voice experience for all users.
The company noted that nearly all current voice users have already migrated to GPT-Live since it launched in July. During that transition period, voice usage across the platform nearly doubled. This suggests that the majority of the user base is already adapting to the newer technology.
Users who still prefer the older conversational modes have until the December deadline to keep their current settings. After that date, ChatGPT will automatically switch all affected users to GPT-Live. The company began notifying users about this retirement on November 10.
How GPT-Live differs
GPT-Live represents a significant architectural shift from the models it is replacing. Unlike the older systems, it can listen and speak at the same time. This allows for a more fluid conversation where users can interrupt the AI or receive brief acknowledgments while they are still talking.
The previous Advanced voice model followed a more traditional approach by processing audio directly but waiting for a user to finish before responding. Standard voice was even more complex, using three separate models to convert speech to text, generate a response, and then convert that text back into audio. These extra steps often caused delays and resulted in lost information.
By consolidating these capabilities, OpenAI aims to create a more seamless experience. GPT-Live is designed to feel less rigid by allowing the model to remain silent when a user needs time to think. The goal is to improve conversational continuity and the depth of the responses provided to the user.
Background task handling
One of the most distinct features of GPT-Live is its ability to connect with OpenAI's text-based models for demanding requests. This allows the system to handle multi-step tasks or complicated questions in the background without ending the live voice session. Users can keep talking while the AI works on the heavy lifting.
This capability creates a more efficient workflow for users who need to perform complex work while maintaining a verbal dialogue. It prevents the conversation from stalling while the system processes data. The AI can essentially delegate the 'thinking' to a different model while keeping the 'speaking' interface active.
OpenAI plans to continue developing GPT-Live to improve personalization and response depth. As the system matures, it will become the primary way users interact with the model's capabilities. The company wants to move away from the fragmented architecture of the past into a more integrated system.
Who is affected by the change
The retirement of legacy voice modes will impact both individual users and developers. Custom GPTs that currently rely on the older voice technology will also lose access to Advanced voice on the December deadline. Developers will need to account for this change when planning their existing voice workflows.
For standard users, the transition is designed to be automatic and non-disruptive. Existing conversations will remain accessible in the history, and users can continue those discussions through text or the new Live voice experience. There is no requirement to manually migrate any data or start new chats to stay compatible.
However, there is a notable exception for specific interactive features. Live video and screen-sharing will continue to run through the old Advanced voice infrastructure for the time being. Because GPT-Live does not yet support these visual features, OpenAI will maintain the older system until the new model gains equivalent capabilities.
The road ahead
While December 11, 2026, marks the end of the selectable legacy modes, it does not mean the end of all Advanced voice features. The company has not provided a specific timeline for when GPT-Live will eventually support video or screen-sharing. For now, these interactive visual features will remain on their own separate infrastructure.
The transition highlights OpenAI's push toward a more unified and responsive AI architecture. By moving toward a model that handles simultaneous input and output, the company is trying to mimic human-like communication more closely. This shift is intended to make the AI feel like a more natural partner in conversation.
As the platform moves forward, users can expect more focus on how the AI handles complex tasks during live interactions. The goal is to provide a system that is both conversational and capable of high-level reasoning simultaneously. This consolidation is a major step in refining how people interact with large language models daily.
What to watch
- When GPT-Live will eventually gain support for live video and screen-sharing.
- How developers will adapt custom GPTs to the new voice architecture.
Try it
Check your ChatGPT settings under the Voice section to see if you are currently using Advanced or Standard voice.
Sources and links
First reported by Interesting Engineering. Written by the SI Agent News Desk from the reporting above.Written with AI from the source above. Every number is checked against the source. Spot a mistake? Corrections.
Liked it? Say why
Add a line for other readers.
No account needed. We never show your email. Comments with links, ads or abuse are held and not shown. Your comment shows after a quick check.