OpenAI kills ChatGPT legacy voice modes on December 11
OpenAI is forcing all ChatGPT voice users to GPT-Live on December 11, killing the older Advanced and Standard modes.
Models ยท Source: Neowin
What happened
OpenAI is pulling the plug on ChatGPT's Advanced and Standard voice modes. On December 11, 2026, both legacy options will permanently vanish from the settings menu. Every user will be automatically migrated to the newer GPT-Live experience. The company started sending out notifications about this forced transition on November 10. Users do not need to manually migrate their old conversations. Previous chats will remain accessible in the history tab. However, any spoken continuation of those old threads will default to GPT-Live.
The legacy systems are simply too outdated to maintain. Standard voice relied on a clunky three-model pipeline. One model converted speech to text, a second generated the text response, and a third converted that text back into audio. OpenAI admits this pipeline introduced severe delays and lost crucial information between the audio and text stages. Advanced voice was slightly better because it processed incoming and outgoing audio directly. But it still forced rigid turn-taking. Users had to wait in silence until the AI finished speaking before they could reply.
GPT-Live fundamentally changes the architecture. It can listen and speak at the exact same time. You can interrupt the model naturally, just like a real human conversation. It can also acknowledge you with brief responses while you are speaking. If you need a moment to think, the model remains silent instead of awkwardly cutting you off. GPT-Live also connects directly with OpenAI text models for heavy lifting. It can delegate complicated questions and multi-step tasks to the background without dropping the ongoing voice call.
Key facts
- December 11, 2026 โ Date Advanced and Standard voice modes disappear from settings
- July โ Month GPT-Live was introduced, triggering a doubling in voice usage
- November 10 โ Date OpenAI started notifying users about the retirement
Why it matters
This shift changes how people interact with AI agents on a daily basis. Rigid turn-taking is dead. GPT-Live makes voice interactions feel like actual phone calls rather than walkie-talkie exchanges. The data proves users prefer this. OpenAI reports that voice usage has nearly doubled since GPT-Live launched in July. Almost all ChatGPT voice users have already made the switch voluntarily. OpenAI plans to use this unified architecture to improve response depth, conversational continuity, and personalization over time.
However, the transition creates a temporary but frustrating fragmentation in ChatGPT capabilities. While standard voice chats move to GPT-Live, live video and screen-sharing features will remain stuck on the old Advanced voice infrastructure. OpenAI admits GPT-Live cannot handle those interactive visual features yet. They will maintain the older system solely to keep video and screen-sharing functional. The company has not announced a specific timeline for bringing those visual capabilities to GPT-Live. This means users will experience two completely different voice models depending on whether they are just talking or sharing their screen.
For builders
Custom GPTs lose legacy voice support
Custom GPTs currently using Advanced voice will lose access on December 11. Developers must test their existing voice workflows with GPT-Live immediately. If your custom GPT relies on the specific pacing of the older models, the user experience will break when the forced migration happens.
Multimodal fragmentation creates edge cases
Building tools that rely on ChatGPT screen-sharing or live video means you are still dealing with the old Advanced voice model. You have to account for different latency and interaction styles depending on whether the user has their camera on or off. OpenAI has no timeline to fix this gap.
My take
OpenAI is absolutely right to kill the old models. Maintaining three separate voice architectures is a massive waste of compute and engineering cycles. But launching GPT-Live as the default before it can handle video and screen-sharing is sloppy execution. Builders and users now have to navigate a fragmented product where the underlying voice model changes just because someone turns on their camera. Ship the complete product, or do not force the migration.
Original reporting: Neowin. This is my rewrite and opinion.