announced a strategic investment and partnership with Arabic.ai. 
LLM-X ranks #1 on Stanford HELM Arabic.
A 91.9% mean score, first of 41 models evaluated, ahead of the leading frontier models. Source: Stanford CRFM HELM Arabic leaderboard, read 6 October 2026.
91.9%
LLM-X mean score on Stanford HELM Arabic
#1 of 41 models
- LLM-XArabic.ai91.9
- Best frontier modelHELM Arabic84.1
41 models evaluated. LLM-X is first.
HELM Arabic Enterprise.
v1.1.0, 25 June 2026, across three task areas.
- Arabic content generation
- Financial reasoning
- UAE legal question answering
| Model | Mean score | Version |
|---|---|---|
 | 0.825 | Arabic.AI LLM-X (2026-06-07) |
 | 0.724 | Arabic.AI LLM-S |
How to read these results: they are benchmark scores. Some tasks use model-based grading, and the scores do not measure production reliability.
Built in the open with Stanford CRFM.
Two Arabic benchmarks, from first publication to the latest results.
HELM Arabic is published.
Stanford CRFM. Yifan Mai, Rida Taleb and Ahmed Elhagry.
Arabic.AI and Stanford CRFM announce their collaboration.
To advance Arabic AI benchmarking.
Read the announcementHELM Arabic Enterprise is published.
Stanford CRFM, developed in collaboration with Arabic.ai. Yifan Mai, Rida Taleb and Ahmed Elhagry.
HELM Arabic Enterprise launches.
Arabic content generation, financial reasoning and UAE legal question answering.
Read the announcementVersion 1.1.0 results.
LLM-X scores 0.825 and LLM-S 0.724, macro-averaged across the three task areas.
LLM-X leads the HELM Arabic leaderboard.
First of 41 models, with a 91.9% mean score, as read on this date.
See the leaderboard
98%
clinician satisfaction in Emergency Department scenarios
Validated by clinicians.
In the HIKMA AI evaluation, blinded, independent panels of clinicians and hospital executives preferred LLM-X over a leading global LLM on every scored dimension.
Read the HIKMA AI storyBuilt on 20 billion words of Arabic.
Arabic.ai builds on Tarjama's legacy: more than 20 billion words of Arabic processed since 2008, across the dialects people actually use.
Explore LLM Models20B+
words processed
22
Arabic dialects supported
1,000+
clients
98%
client retention rate
Arabic as spoken in the Gulf, Egypt, the Levant, Iraq, the Maghreb and Sudan
We measure for Arabic understanding1, task specialization2 and deployability3.
- 1Arabic understanding
Professional Arabic involves register, terminology and context. Evaluate the material your teams actually read and write.
- 2Task specialization
Drafting, recurring processing and translation need different acceptance criteria.
- 3Deployability
A useful model must fit the information boundary, workload and review process around it.
Make your next ambition operational.
Bring intelligence into your knowledge, people and operations. Talk to our team about the work you want to transform.