OpenAI has embarked on a new chapter by releasing IndQA, a groundbreaking benchmark crafted to deepen AI’s understanding of India’s rich languages and cultures. Alongside IndQA, OpenAI introduced the Teen Safety Blueprint—a step toward making AI safer for young people. To sustain and accelerate this progress, OpenAI also secured a $38 billion, seven-year partnership with Amazon Web Services (AWS), giving its AI projects the powerful technological foundation they need for the years ahead.
IndQA: AI Learns from India’s Diversity
IndQA—short for Indian Question-Answering benchmark—was launched on November 4, 2025. It is designed to test how well AI systems understand and reason through questions that are deeply rooted in Indian languages, culture, and everyday life.
- 2,278 questions crafted across 12 Indian languages: Hindi, Hinglish (blending Hindi and English), Gujarati, Punjabi, Kannada, Odia, Marathi, Malayalam, Tamil, Bengali, Telugu, and English.
- The questions span 10 cultural domains including architecture, arts, everyday life, law, media, religion, sports, literature, food, and history.
- Questions are natively written in each language—never translated—ensuring authenticity, proper use of local phrases, and meaningful context.
- This benchmark was built with the expertise of 261 domain experts from across India, guaranteeing accuracy and cultural respect.
- Each AI answer is evaluated through rubric-based grading, much like responses in a well-judged exam—measured by strict and clear standards to capture subtle strengths and weaknesses in understanding.
Previous global AI benchmarks too often focused on English or on translations that missed local meaning. IndQA addresses this gap, serving as a guiding standard for anyone seeking to build AI that truly understands India’s unique blend of languages and cultures. IndQA sets a precedent: it moves AI closer to people’s daily realities in India, fostering more relevant and useful technology for millions.
How AI Performs on IndQA
The first assessments show that even the most advanced AI models face real challenges understanding India’s diversity:
- GPT-5 (Thinking High) leads among the tested models, reaching about 34.9% accuracy on average. It performs strongest in Hindi and Hinglish, scoring near 45% and 44%, but struggles much more in Bengali and Telugu.
- Other leading AI systems, such as Gemini 2.5 Pro and GPT-4o, score lower still, revealing that Indian language understanding is not a solved issue, even for top technology.
IndQA is not meant to pit languages or models against each other, but to guide steady progress in each language over time, revealing new strengths and weaknesses as AI evolves.
Teen Safety Blueprint: Protecting Young AI Users
Safety remains a core concern as AI tools become more serious parts of daily life. OpenAI’s Teen Safety Blueprint introduces several protections:
- Parental controls to allow guardians to set limits and supervise how their teens interact with AI.
- AI-powered age prediction that helps detect user age ranges, adapting experiences to keep younger audiences safe.
This blueprint embodies OpenAI’s commitment to balancing broad access to its tools with strong and thoughtful protections for younger users.
Building an AI Future with AWS
To support rapid advances in technology, OpenAI has secured a transformative $38 billion, seven-year partnership with AWS. This agreement ensures the vast computing power necessary to train, deploy, and maintain larger, smarter AI models—like the new GPT-5.1, designed with advanced reasoning skills. This partnership underscores OpenAI’s ambition to serve massive user bases and drive research breakthroughs, including those tailored for intricate markets like India.
A Step Forward for AI and for India
IndQA offers more than numbers. It signals a respect for India’s intricate tapestry of voices and stories. Coupled with stronger safety measures and the infrastructure to power future models, OpenAI’s latest initiatives mark a meaningful stride toward AI that is not only more capable, but also more inclusive and responsible.
Leave a Reply