IndicKLAR (Urdu): leaderboard
Factual knowledge recall in Urdu from IndicKLAR (Mazumder, Pathak, Kodali, Joshi, Agarwal and Patro, 2026): 2,619 KLAR-CLC queries about entities across 20 relations, machine-translated without native-speaker review. Accuracy in percent counts an answer when it and the gold answer are case-insensitive prefixes of each other (higher is better), with direct 3-shot prompting using English examples and greedy decoding.
Metric: Accuracy (%; relaxed prefix match of the answer to 2,619 factual queries in Urdu script, machine-translated, not speaker-verified; direct 3-shot prompting with English examples, greedy decoding). Source: arxiv.org. Saturation forecast: Estimated already saturated. 9 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemma 3 12B (IT) | 68.8 | #1849 |
| 2 | Qwen 2.5 14B Instruct | 59 | #1745 |
| 3 | Llama 3.1 8B Instruct | 57 | #4277 |
| 4 | Qwen 2.5 7B Instruct | 48.5 | #2583 |
| 5 | Gemma 3 4B (IT) | 48.3 | #3848 |
| 6 | Llama 3.2 3B Instruct | 40.7 | #7185 |
| 7 | Llama 3.2 1B Instruct | 26.5 | #9109 |
| 8 | Qwen 2.5 1.5B Instruct | 25.1 | #7518 |
| 9 | Gemma 3 1B (IT) | 13.7 | #8752 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=indicklar-urdu · How It Works · Data refreshed daily, snapshot 2026-11-12.