Research · Open Arabic medical language model
Hakeem-7B: an open Arabic medical reasoning model, built in Cairo
Open Arabic medical models sit at 40–50% on MedAraBench, below the level where output can be trusted. We trained a 7-billion-parameter model that reasons in Arabic and crossed the 60% line, then published the model and the method.
- Size
- 7 billion parameters
- Base model
- Falcon-H1-7B
- Domain adaptation
- 14,404 medical documents
- Supervised fine-tuning
- 9,714 reasoning pairs
- Preference tuning
- 387 pairs
- Peer-reviewed in
- Knowledge-Based Systems, Elsevier, 2026
MedAraBench clean split, think-first mode, 2026.
Where it stands
On a 14-model shared-parser leaderboard, Hakeem-7B ranks above every open model in the 8B-and-under class and above OpenBioLLM-70B and Llama-3.1-70B, and is statistically level with Gemma-2-27B and Med42-70B.
Limits
This is a research model, published openly with its method. It is not for diagnosis, triage, dosing or treatment decisions, and its output needs verification by a qualified clinician. We keep research models separate from the production systems hospitals run.
Updated