Pak3H - PakAlpaca (Urdu): leaderboard
The helpfulness part of Pak3H (2026): 4,445 Alpaca instructions translated into Urdu and adapted to Pakistani contexts by native annotators, with each model response compared against a language-matched reference response by a GPT-4o-mini judge in the AlpacaEval pairwise format. Win rate in percent; higher is better.
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 March 2028. 3 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Llama 3 70B Instruct | 36.2 | #1676 |
| 2 | DeepSeek V3.2 | 29.26 | #406 |
| 3 | GPT-4o Mini | 13.19 | #1623 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=pak3h-pakalpaca-urdu · How It Works · Data refreshed daily, snapshot 2026-11-13.