Kaggle IndicGenBench: leaderboard
Google DeepMind's IndicGenBench on Kaggle: text generation in up to 29 Indic languages across cross-lingual summarization, translation and two question-answering tasks. This board is Kaggle's unweighted average of the four task scores, kept only for models with all four; it mixes ChrF and token-F1 points (0-100, higher is better).
Metric: Average Score. Source: www.kaggle.com. Saturation forecast: Around June 2027. 12 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Sonnet 4.5 | 60.65 | #148 |
| 2 | Gemini 3 Pro (Preview) | 59.89 | #71 |
| 3 | Gemini 2.5 Flash | 56.92 | #255 |
| 4 | Claude Opus 4.1 | 54.96 | #146 |
| 5 | GPT-5 | 52.23 | #105 |
| 6 | Grok 4 | 52.03 | #189 |
| 7 | O3 | 49.97 | #132 |
| 8 | Qwen 3 235B A22B 2507 Instruct | 48.91 | #315 |
| 9 | GPT-4.1 | 43.66 | #258 |
| 10 | O4 Mini | 41.85 | #185 |
| 11 | GPT-OSS-120B | 39.12 | #346 |
| 12 | Sarvam M | 37.16 | #643 |
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 · How It Works · Data refreshed daily, snapshot 2026-11-05.