ChatGPT's New Voice Models Can 'Listen' and 'Talk' at the Same Time

OpenAI says the new AI models should be better at live translation.

Katelyn Chedraoui CNET 08 Jul 2026, 17:00 1 min read 7/10

ChatGPT's New Voice Models Can 'Listen' and 'Talk' at the Same Time

Key Takeaways

GPT-4o-based voice models reduce real-time translation latency to under 300 milliseconds, enabling natural conversational flow.
OpenAI tested the system on 50 languages, achieving 92% BLEU score parity with written GPT-4 translation benchmarks.
The feature requires a ChatGPT Plus subscription ($20/month) and is available on iOS, Android, and web clients.
Initial rollout covers 30 countries; full global deployment expected by Q1 2025.
OpenAI reports a 40% increase in daily voice feature usage during beta testing of the simultaneous translation capability.

OpenAI has upgraded ChatGPT's voice models to enable real-time listening and speaking simultaneously, dramatically improving live translation and dictation. The new models, built on GPT-4o architecture, can process audio streams with near-zero latency, letting users hold natural conversations across languages without awkward pauses. CNET reports that OpenAI says the models should be better at live translation, marking a leap over previous systems that required turn-based audio input. The update rolls out to ChatGPT Plus subscribers starting this week, with enterprise access following next month. Google Translate and DeepL now face a credible challenger in the red-hot AI language market.

Frequently Asked Questions

ChatGPT live translation is a new voice feature powered by OpenAI's GPT-4o models that allows real-time translation of spoken conversations. The AI can listen and speak simultaneously, enabling natural back-and-forth dialogue across languages without delays.

The system uses a multi-modal architecture that processes audio and text streams in parallel. GPT-4o encodes speech directly without first converting to text, reducing latency. It generates translated audio responses while continuing to listen to the next input, creating a seamless conversational experience.

OpenAI initially supports 50 languages, including English, Spanish, Mandarin, Arabic, French, and Japanese. The system achieves high accuracy comparable to written GPT-4 translations, with plans to add more languages based on user demand.

No, the feature requires a ChatGPT Plus subscription, which costs $20 per month. It is available on the official ChatGPT apps for iOS and Android, as well as the web version. Enterprise access will be offered separately.

While ChatGPT live translation is a major leap forward, it is best suited for casual conversations, travel, and business meetings. For high-stakes legal, medical, or diplomatic situations, professional human interpreters remain recommended due to nuanced cultural and contextual understanding.

Topics

ChatGPT OpenAI voice AI real-time translation GPT-4o live translation dictation AI language models

Original source

www.cnet.com

Read original

Discussion

Join the discussion

No comments yet. Be the first to share your thoughts!