OpenAI announced on October 15 the start of the deployment of GPT-Live-1.1 and GPT-Live-1.1 Quip, a smaller version of the model, in ChatGPT Voice. The company describes improvements in the naturalness of pronunciations in different languages and the addition of voices to choose from. This is a deployment announcement, so the appearance of the options on one user does not guarantee that they are already available on every account and every region.
According to the update log, the model also uses a wider variety of short sentences to indicate that it is looking for information. In terms of conversation, these signals can help us understand when something's happening in the background. They are not evidence that every answer involves searching, and do not replace source checking when the question concerns an important or changing detail.
Which voices were added and how to select them
OpenAI notes new voices for French, German, Hindi, Japanese, Korean and Mexican Spanish, as well as a voice named Bubbie in English. The list does not include a notification of a new Hebrew voice. A general improvement in accents is also insufficient to establish changes in Hebrew quality, availability in Israeli accounts or pronunciation of local names.
To select a voice, the company refers to the ChatGPT settings, to Voice and from there to the voice selection. If a certain option does not appear, it is worth considering that the availability may vary by account and region. The announcement offers no basis for guaranteeing that another setting will unlock a missing voice, or a single date when all users will receive the same list.
What to look for in a real conversation
Anyone who uses voice for work meetings, language practice, or understanding documents can try a short clip that represents their use. You should pay attention to names, numbers and transitions between languages. A natural-sounding voice can make listening easier, but the comfort of presentation and accuracy of content are two things to check separately. A fluent spoken response can still contain an error.
A useful test is to ask for pre-known information and check if the details are preserved, then add a question requiring clarification or correction. That way you can feel how the conversation is going without relying on impressions from the first sentence alone. This assesses suitability for an individual's use; it is not a comparative test proving the model is better than every alternative or in every language.
The announcement did not change prices or provide details establishing new usage limits for each plan. Therefore, users do not have to change their subscription just because of the new model names. The practical change concerns the voice model and the selection of voices now being rolled out. Before relying on it on a regular basis, you should check what actually appears on the account and what suits your needs for listening and accuracy.