indic-gpqa

GPQA is a multiple-choice dataset of very hard, "Google-proof" questions in biology, physics and chemistry, written and validated by domain experts: experts with or pursuing PhDs in the field reach 65% accuracy, while skilled non-experts with unrestricted web access reach 34%. This dataset translates the 198-question Diamond subset.

indic-gpqa is GPQA Diamond translated into 14 Indian languages by an LLM translation pipeline with automatic quality checks, with a portion of the translations reviewed and corrected by human annotators. It is part of the Indic-LLM-Eval-Suite.

Upstream source: Idavidrein/gpqa

Languages

Assamese (as), Bengali (bn), Gujarati (gu), Hindi (hi), Kannada (kn), Malayalam (ml), Marathi (mr), Nepali (ne), Odia (or), Punjabi (pa), Sanskrit (sa), Tamil (ta), Telugu (te), Urdu (ur)

Layout

One row per original benchmark item, with every language on that same row. The original columns are carried through unchanged; each translated field gains one column per language:

Question -> Question_Hindi_translation, Question_Bengali_translation, ...

rows	198
languages	14
translated fields	Question, Correct Answer, Incorrect Answer 1, Incorrect Answer 2, Incorrect Answer 3
columns	148

More details

For how the translations were made and verified, and for model results on the suite, see the report and the blog:

• Report: coming soon
• Blog: Introducing Indic-LLM-Eval-Suite

Citation

If you use this dataset, please cite:

@misc{indicllmevalsuite2026blog,
  title        = {Introducing Indic-LLM-Eval-Suite: Better Indic Models Need Better Benchmarks},
  author       = {Kunchukuttan, Anoop and Kamath, Anusha and Khapra, Mitesh M. and
                  Khan, Mohammed Safi Ur Rahman and Agarwal, Niket and Thethy, Rajpreet and
                  Akkati, Ramani and Joshi, Raviraj and Murthy, Rudra and
                  Jayakumar, Thanmay and Vaidya, Utkarsh},
  year         = {2026},
  month        = oct,
  howpublished = {\url{https://ai4bharat.github.io/indic-llm-eval-suite-blog/}},
  note         = {AI4Bharat, IIT Madras and NVIDIA}
}