The way we talk to AI is changing. OpenAI is rolling out significant updates to ChatGPT's voice capabilities, aiming to make conversations feel less robotic and more human, particularly for its vast user base in non-English speaking regions.
New Voices, Smarter Conversations
OpenAI has
begun rolling out GPT-Live-1.1, an update to its voice model that introduces 13 new voice options across several languages, including Hindi, French, German, Japanese, Korean, and Mexican Spanish. For users in India, this update introduces two specific Hindi voices named 'Neer' and 'Leher', designed to make interactions feel more natural and familiar for Hindi speakers. The rollout, which began in mid-October 2026, aims to improve pronunciation and create more natural-sounding accents across languages. This is part of a broader strategy to move beyond a one-size-fits-all approach and cater to a global audience. The new voice model is an evolution of the GPT-Live architecture introduced in July 2026, which allows the AI to listen and speak simultaneously—a 'full-duplex' system that mimics real human conversation. This means users can interrupt, pause to think, and the AI will offer conversational acknowledgements like 'mhmm' or 'yeah' to show it’s following along.
A Strategic Push for India
This focus on localization is no accident; it is a calculated business move. India is one of OpenAI's largest and most strategic markets, and the company recognizes that text-based input is not always the most comfortable or efficient method for many users. Typing in Indic scripts on mobile keyboards can be a barrier to entry. By prioritising voice, OpenAI is adapting to how many Indians already communicate, often using voice notes on platforms like WhatsApp. The company's leadership has stated that for AI to be truly useful, it must be accessible regardless of language or occupation. The recent updates specifically target the nuances of the Indian linguistic landscape, including the common practice of 'code-switching' between English and regional languages in a single conversation. This investment in India-specific features serves as a template for how the company might approach other multilingual, mobile-first markets around the world.
Beyond Translation to True Understanding
The latest updates are about more than just adding new languages; they represent a fundamental shift in how AI understands and processes speech. A recent upgrade quietly enabled ChatGPT's microphone to understand over 70 languages and even mix them within the same sentence without needing to manually switch settings. This is a significant leap for multilingual speakers who naturally blend languages in daily conversation. The underlying models have also become more powerful. The voice system can delegate complex queries that require deep reasoning or web searches to more advanced models like GPT-5.5 or higher, all while maintaining the flow of conversation. This seamless handover feels more natural in the latest version, which also reduces the repetitive 'checking' or 'looking that up' phrases that could make conversations feel stilted. It’s a move from simple speech-to-text and back again, towards a system that has a more holistic, context_aware grasp of a spoken request.
How to Access the New Features
The new voice models are rolling out globally to ChatGPT users on iOS, Android, and the web. The specific model you access depends on your subscription plan. Free users will generally get a 'mini' or 'Quip' version of the voice model, while paid subscribers on plans like Plus and Pro get access to the more powerful versions with higher usage limits. To use the new features, users can tap the voice (waveform) icon in the ChatGPT app. In the settings menu, you can select your preferred voice and primary language to help the AI better understand you. For the new Hindi voices, 'Neer' and 'Leher', eligible users in India can select them in the voice settings to begin conversing in Hindi. Users with access to the premium models can also adjust the 'intelligence' level, choosing between faster responses or allowing the AI more time to think for more complex answers.
