ArabCulture-Dialogue - Dialect Steering: leaderboard

ArabCulture-Dialogue (MBZUAI, 2026) tests dialect steering: given a dialogue context and a Modern Standard Arabic utterance, the model writes the next turn in the dialect of a named Arab country. The score is the percent of continuations whose GlotLID language code exactly matches the target dialect; higher is better.

Metric: Dialect identity accuracy (%): share of continuations whose GlotLID language code exactly matches the target country's dialect code (strict ISO 639-3), zero-shot dialect steering: given a dialogue context and an MSA utterance, the model writes one continuation in the target country's dialect, over 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); higher is better. Source: arxiv.org. Saturation forecast: Around February 2027. 12 models tracked. Updated 7 Oct 2026.

Top models

#ModelScoreOverall rank
1Gemini 2.5 Pro50.5#276
2GPT-545.4#178
3ALLaM-7B-Instruct-preview36.25#2001
4Jais-2-8B-Chat20.8#2496
5c4ai-command-r7B-arabic-02-202518.6#2135
6Gemma 2 9B (IT)16.4#2369
7SILMA-9B-Instruct-v1.07.8#3847
8Llama 3.1 8B Instruct6.4#4600
9Qwen 3 8B4.1#1873
10Fanar-1-9B3.8#1913
11jais-adapted-7B-chat2.2#5831
12Hala-9B1.6#1162

No result here: #1 Claude Fable 5.2, #3 Gemini 4 Neon, #4 GPT-6.2 Astra.

Interactive version: theaggregate.ai/benchmark?slug=arabculture-dialogue-dialect-steering · How It Works · Data refreshed daily, snapshot 2026-11-21.