OpenAI rolled out GPT‑Live, a voice‑capable generation of its models, to ChatGPT users today, and the interface now defaults to playing audio responses – you’ll want to turn the sound on to hear it OpenAI tweet.
How GPT‑Live works
GPT‑Live streams audio directly from the model, converting text to speech in real time and accepting spoken prompts via the microphone. The pipeline runs on OpenAI’s servers, so latency is comparable to normal ChatGPT replies, but the extra audio processing adds a few hundred milliseconds.
What this means for shipping engineers
Because the feature is currently limited to the ChatGPT web and mobile UI, there’s no public API yet, so you can’t embed it in a product today. However, the rollout signals that OpenAI may soon expose a voice endpoint, which could replace custom TTS/ASR stacks in low‑budget startups.
Caveats and cost considerations
The voice model is bundled into the existing ChatGPT experience, so there’s no separate price tag, but audio generation will count against your usage limits and could accelerate quota exhaustion. Accuracy varies with background noise, and early users report occasional mis‑recognition that requires manual correction.
When to try it
If your team is prototyping conversational interfaces, enable sound in ChatGPT now to evaluate latency and quality before a potential API release. Keep an eye on OpenAI announcements for a developer preview and pricing details.