ArabCulture-Dialogue - MSA to Dialect Translation (BLEU): leaderboard
ArabCulture-Dialogue (MBZUAI, 2026) asks models to translate culturally grounded dialogues from Modern Standard Arabic into the dialect of a named Arab country, zero-shot with the target country and region named in the prompt. The score is BLEU-4 against the human reference translation on a 0-100 scale; higher is better.
Metric: BLEU-4 (0-100, the paper's 0-1 score times 100) against the human reference, zero-shot MSA to Dialect translation of the dialogues of ArabCulture-Dialogue (13 Arab countries, parallel Modern Standard Arabic and country-dialect versions of 3,471 culturally grounded multi-turn dialogues built from ArabCulture), with the target country and region named in the prompt; higher is better. Source: arxiv.org. Saturation forecast: Around March 2028. 11 models tracked. Updated 7 Oct 2026.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5 | 27.6 | #185 |
| 2 | Gemini 2.5 Pro | 27.3 | #280 |
| 3 | ALLaM-7B-Instruct-preview | 19.6 | #1992 |
| 4 | Fanar-1-9B | 15.6 | #1925 |
| 5 | c4ai-command-r7B-arabic-02-2025 | 15.2 | #2140 |
| 6 | jais-adapted-7B-chat | 12.9 | #5804 |
| 7 | Hala-9B | 12.2 | #1167 |
| 8 | Qwen 3 8B | 11.5 | #1890 |
| 9 | Gemma 2 9B (IT) | 7.1 | #2383 |
| 10 | Llama 3.1 8B Instruct | 5.8 | #4655 |
| 11 | SILMA-9B-Instruct-v1.0 | 3.5 | #3851 |
No result here: #2 Claude Fable 5.2, #3 Gemini 4 Neon, #4 GPT-6.2 Astra.
Interactive version: theaggregate.ai/benchmark?slug=arabculture-dialogue-msa-to-dialect-translation-bleu · How It Works · Data refreshed daily, snapshot 2026-11-22.