OpenAI’s IndQA Benchmark Puts Indian Languages & Culture in Focus for AI

November 2025 — OpenAI has introduced a benchmark named IndQA, aimed at testing how well AI systems understand questions rooted in Indian languages and cultural contexts. The dataset comprises 2,278 questions, written natively in 12 Indian languages, including Hindi, Tamil, Telugu, Bengali, Gujarati, Odia, and Hinglish.

It covers ten domains such as food & cuisine, law & ethics, literature & linguistics, and arts & culture. An important thing to notice here is that the questions aren’t simply translations of English-based prompts; they were crafted by 261 native experts and filtered to preserve those that current top models struggle with.

There are several reasons behind the release of IndQA.

First, it directly targets a gap in existing AI benchmarks. Most benchmarks focus on translation or multiple-choice tasks and are saturated at the top end, making further progress difficult to measure.

Second, it highlights that AI cannot be globally useful if it relies solely on English and Western culture as its foundational context. It must navigate linguistic diversity and cultural nuance. Given that India is a massive and rapidly growing market for AI tools (and one of ChatGPT’s biggest markets second only to the U.S.), this move makes strategic sense.

Looking ahead, IndQA could shift how AI models are evaluated and trained worldwide. By emphasising culturally grounded reasoning in multiple languages, it sets a precedent.

Future models may increasingly need to be literate not just in languages but in local customs, histories, and everyday life. That means better localisation of AI for millions who speak non-English languages, deeper inclusion of regionally diverse content, and likely more competition to build models that understand more than just standard globalised datasets.

For India in particular, this could lead to AI tools that are genuinely tailored, linguistically, culturally, and contextually, to the country’s many communities.