OpenAI launches GPT-Live voice models for more natural conversations

OpenAI has introduced GPT-Live, a new family of voice models that can speak and listen simultaneously, aiming to make conversations with ChatGPT feel more like talking to a human. The models, GPT-Live-1 (for paid subscribers) and GPT-Live-1 mini (for free users), are rolling out globally from July 8, 2026.

What's new

Unlike previous turn-based voice modes, GPT-Live uses a full-duplex architecture that processes input and output continuously Source: The Verge. This allows users to interrupt naturally, and the model can interject with acknowledgments like “mhmm” or stay silent when needed. It also delegates complex tasks—such as web search or reasoning—to OpenAI’s GPT-5.5 in the background, keeping the conversation flowing Source: TechCrunch.

Reception and concerns

Early tests by users highlight persistent weaknesses. TikToker @huskistaken asked GPT-Live to count the letter E in “seventeen”; the model answered “two” instead of the correct “three” Source: Gizmodo. Another user reported the model struggling with the classic “strawberry” R-count test Source: Gizmodo. These failures underscore that while the model improves conversational flow, its reasoning on simple tasks can still falter.

Some reviewers praised the model’s natural feel. ZDNet found that “ChatGPT can now speak with you almost as if it were a real person” Source: ZDNet. However, a Gizmodo article noted a new annoyance: the model’s habit of interrupting with affirmative noises, calling it an “interrupting machine” Source: Gizmodo.

Safety and availability

OpenAI added safeguards to steer harmful responses, provide crisis resources, and offer age-appropriate replies for teens Source: Engadget. The model is available on iOS, Android, and the web. Paid users get GPT-Live-1, while free users default to the smaller mini variant.