Arabic.AI's LLM-X topped HELM Arabic with a mean score of 0.86, ahead of Gemini 2.5 Flash at 0.82

What the chart shows

HELM Arabic, built by Stanford’s Center for Research on Foundation Models with Arabic.ai, averages model performance across seven Arabic-language benchmarks covering academic assessment, grammar and region-specific safety. Of the 29 models evaluated, Arabic.AI’s LLM-X, a regionally developed model, scored highest at 0.86, ahead of Gemini 2.5 Flash (0.82) and GPT-5.1 (0.81).

The 15 models the AI Index groups under 2025 scored between 0.69 and 0.86. Among the 14 grouped under 2024, Arabic-focused models such as Jais and AceGPT ranged from 0.53 to 0.74, and Falcon3-7B-Instruct scored lowest at 0.37.

Key numbers

How to read it

• The score is HELM’s mean across seven Arabic-language benchmarks, from 0 to 1.
• The AI Index groups the models into 2024 and 2025; the year is in the data file.
• Results apply to these 29 models on these benchmarks. The AI Index notes that rankings from English-centric evaluations do not necessarily hold when benchmarks reflect local usage, dialect and cultural references.