HiMed (Hindi): leaderboard
Hindi medical reasoning suite from HiMed (Jiang et al., 2026): the mean accuracy over five Hindi multiple-choice sets, namely lexicon-guided Hindi translations of MMLU-Pro Biology, MMLU-Pro Health and GPQA-med, the Hindi NEET-UG questions of HiMed-West Exam and the HiMed-Trad questions on Indian systems of medicine. Each set is zero-shot multiple choice scored with a relaxed answer extractor, averaged over three runs; accuracy in percent, higher is better. In the released files every GPQA-med answer is option D, so that fifth of the average also rewards a bias toward the last option.
Metric: Accuracy (%; mean of the five Hindi sets, zero-shot multiple choice, mean of three runs). Source: arxiv.org. Saturation forecast: Estimated already saturated. 11 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-4o | 54.7 | #741 |
| 2 | Qwen 2.5 14B Instruct | 40.7 | #1745 |
| 3 | Qwen 2.5 7B Instruct | 35.3 | #2583 |
| 4 | Llama 3.1 8B Instruct | 34.2 | #4277 |
| 5 | aya-23-8B | 26.5 | #6778 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=himed-hindi · How It Works · Data refreshed daily, snapshot 2026-11-12.