OpenAI to retire ChatGPT's Advanced and Standard voice modes

If you still rely on ChatGPT's older voices for your daily commutes or study sessions, your settings menu is about to get a lot simpler.

OpenAI is officially retiring both Advanced voice and Standard voice on December 11, moving all spoken chats exclusively to GPT-Live. The rollout of notifications started on November 10. While your old chat history will stay untouched, any new spoken follow-ups after the December cutoff will automatically run on the newer model.

There is one small exception: live video and screen sharing will still lean on Advanced voice behind the scenes, at least until GPT-Live picks up native visual support.

Ditching turn-based chatter for true full-duplex

OpenAI isn't cutting these options just to tidy up the settings screen. Under the hood, the way voice is processed has fundamentally evolved.

Standard voice was basically a Rube Goldberg machine of three separate systems: one transcribed your audio to text, a second generated an answer, and a third read the text back to you. OpenAI notes that this chain introduced obvious latency and stripped out subtle nuances.

Advanced voice improved on this by natively processing raw audio, but it was still strictly turn-based. You spoke, you paused, and only then did the model start generating a response.

GPT-Live, by contrast, is full-duplex. It can listen and speak simultaneously, tossing in casual backchannel cues like "mhmm" or "yeah" while you ramble, waiting patiently while you pause to think, and delegating heavy background tasks to OpenAI's dedicated text models without pausing the vocal flow. According to OpenAI, overall voice usage has nearly doubled since GPT-Live rolled out in July, with virtually all active voice users already relying on it.

When upgrades clash with user attachments

Even so, cutting legacy options is rarely painless when users form personal bonds with synthetic personas.

Over on the OpenAI developer forum, users have already started pleading with the company to preserve the original Standard voice, especially legacy tones like Breeze. One poster described the voice as a daily source of emotional comfort and argued that GPT-Live, despite its real-time speed, feels overly artificial and lacks the gentle pacing of the older voice.

OpenAI acknowledged in its notice that adapting to new conversation styles can be jarring. However, the company made it clear that the December 11 deadline is final. The team plans to funnel feedback from both retired modes into GPT-Live to improve answer depth, conversational memory, and personalization.

Why it matters

Consolidating the voice stack is a smart, necessary move for OpenAI's operational efficiency.

Maintaining three separate voice pipelines—especially the fragile, multi-hop architecture behind Standard voice—drains engineering cycles and compute budgets. Standardizing on GPT-Live lets OpenAI pour all its optimization efforts into a single, highly adaptable multimodal engine that rivals can't easily replicate with off-the-shelf TTS and ASR components.

At the same time, this retirement highlights an emerging dilemma in consumer AI. Unlike a routine UI redesign, modifying an AI's voice disrupts genuine human attachment. When people lean on a voice for comfort, replacing it with a faster, technically superior model can still feel like a step backward to the person listening.

For the broader industry, the message is loud and clear: turn-based voice bots are officially obsolete. If your conversational AI can't handle real-time interruptions and listen while it talks, you're playing yesterday's game.