IndicIFEval-Ground - Urdu: leaderboard
Metric: Prompt-level loose accuracy (%, 0-100) on the Urdu prompts; IndicIFEval-Ground: prompts generated by Gemini 2.5 from native Indic web text (sangraha-verified) for six constraint types (keyword inclusion, prohibition and frequency, paragraph and sentence count, first word), verified by native speakers; a different prompt set per language; IFEval-style verifiable instructions checked by the IFEval rules with Indic NLP Library tokenization and sentence splitting; prompt-level loose accuracy (a prompt counts when all of its instructions are followed, after minor formatting relaxations); greedy decoding through the LM Evaluation Harness; higher is better. Source: arxiv.org. Saturation forecast: Around 2029. 19 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5 Mini (2025-08-07) | 73.6 | #165 |
| 2 | GPT-5 | 72.7 | #91 |
| 3 | Llama 3.2 3B Instruct | 61.7 | #1321 |
| 4 | Qwen 3 8B | 60.2 | #667 |
| 5 | Gemini 3 Pro | 60.2 | #77 |
| 6 | Qwen 3 14B | 60.2 | #524 |
| 7 | Qwen 3 4B | 59.9 | #823 |
| 8 | Gemini 3 Flash | 59.6 | #93 |
| 9 | Llama 3.2 1B Instruct | 59.4 | #1518 |
| 10 | Qwen 3 32B | 57.6 | #424 |
| 11 | Llama 3.1 8B Instruct | 56.4 | #1018 |
| 12 | Gemma 3 27B (IT) | 54.3 | #509 |
| 13 | Gemma 3 4B (IT) | 54 | #971 |
| 14 | Qwen 3 1.7B | 54 | #1186 |
| 15 | Gemma 3 12B (IT) | 53.7 | #655 |
No result here: #3 Claude Opus 5.5, #5 GPT-6 Astra, #8 Claude Fable 5.1.
Interactive version: theaggregate.ai/benchmark?slug=indicifeval-ground-urdu · How It Works · Data refreshed daily, snapshot 2026-10-11.