OpenAI unveiled IndQA, a benchmark designed to test how well AI systems understand and reason about questions rooted in Indian culture and everyday life—moving beyond translation-style tests. Official post: Introducing IndQA.

• Scope: 2,278 questions spanning 12 languages and 10 cultural domains (e.g., Food & Cuisine, History, Law & Ethics, Literature, Religion, Sports).
• Languages: Bengali, English, Hindi, Hinglish, Kannada, Marathi, Odia, Telugu, Gujarati, Malayalam, Punjabi, and Tamil.
• Authorship: Built with 261 Indian domain experts; each item includes an expert rubric and ideal answer for grading.
• Evaluation method: Rubric-based scoring focused on culturally grounded reasoning; questions filtered adversarially against strong models to preserve headroom.
• Context: Addresses gaps where multilingual benchmarks are saturated and often miss culture-specific reasoning.

Why it matters: As AI adoption grows outside English-first contexts, culture-aware benchmarks like IndQA provide a clearer picture of real capability—and a north star to improve model quality for India’s diverse language ecosystem.