Introducing IndQA

Source: OpenAI Date: 2025-11-03 URL: https://openai.com/index/introducing-indqa

Summary

OpenAI research post from November 2025 introducing IndQA, a question-answering benchmark for Indian languages — likely covering Hindi, Bengali, Tamil, Telugu, and other major Indian language families. The benchmark evaluates model performance on reading comprehension and factual question answering in Indian languages, where training data is significantly sparser than English. Published the same week as the GPT-5.1 developer launch, suggesting a coordinated multilingual capability demonstration.

Implications

Multilingual benchmark as market signal. India is OpenAI’s largest non-English-speaking target market and a significant growth geography. Publishing an Indian language QA benchmark signals both research investment in this area and a commitment to measuring and improving performance in these languages specifically. It’s also a resource for the broader NLP community to evaluate models on these languages.

The data scarcity problem in Indian languages. There are 22 official languages in India and several hundred spoken languages. Training data quality varies enormously — Hindi is relatively well-resourced by global standards; many other Indian languages are severely underrepresented. IndQA’s existence implies OpenAI is measuring the capability gap; whether they’re closing it is a separate question.

Thread: OpenAI India expansion. Sits alongside the OpenAI for India launch (February 2026) and the data residency Asia post (May 2025) as markers of OpenAI’s systematic India market build.

Watch: Whether IndQA scores for GPT-5.x family show meaningful improvement over GPT-4o in Indian languages, and whether the benchmark gets adopted by third parties to evaluate competing models on Indian language performance.