Kaggle IndicGenBench - XQuAD-In: leaderboard
XQuAD-In from Google DeepMind's IndicGenBench on Kaggle: extract a short answer span from a passage for a question in the same Indic language (12 languages). Scored with SQuAD-style token F1 as in the IndicGenBench paper, 0-100, higher is better.
Metric: Token F1. Source: www.kaggle.com. Saturation forecast: Around February 2027. 13 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Sonnet 4.5 | 79.96 | #148 |
| 2 | Gemini 3 Pro (Preview) | 77.68 | #71 |
| 3 | Gemini 2.5 Pro | 77.14 | #159 |
| 4 | Gemini 2.5 Flash | 76.8 | #255 |
| 5 | Claude Opus 4.1 | 72.3 | #146 |
| 6 | Grok 4 | 71.62 | #189 |
| 7 | GPT-5 | 69.01 | #105 |
| 8 | Qwen 3 235B A22B 2507 Instruct | 63.81 | #315 |
| 9 | O3 | 62.4 | #132 |
| 10 | GPT-4.1 | 52.86 | #258 |
| 11 | Sarvam M | 51.48 | #643 |
| 12 | O4 Mini | 50.43 | #185 |
| 13 | GPT-OSS-120B | 50.14 | #346 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=kaggle-indicgenbench-xquad-in · How It Works · Data refreshed daily, snapshot 2026-11-05.