Pak3H - PakAlpaca (Urdu): leaderboard
Metric: Win rate (%; AlpacaEval pairwise preference of the model response over the language-matched reference response on all 4,445 human-contextualized Urdu PakAlpaca instructions, judged by GPT-4o-mini; temperature 0.7, 2,048 new tokens). Source: arxiv.org. Saturation forecast: Around August 2028. 3 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Llama 3 70B Instruct | 36.2 |
| 2 | DeepSeek V3.2 | 29.26 |
| 3 | GPT-4o Mini | 13.19 |
Interactive version: theaggregate.ai/benchmark?slug=pak3h-pakalpaca-urdu · How It Works · Data refreshed daily, snapshot 2026-09-29.