humain-ai/ALLaM-7B-Instruct-preview
Text Generation • 7B • Updated • 9.58k • 208
Open-weight models AI Arm evaluates on Arabic document tasks: cited Q&A, long-document reasoning, report drafting. Curated, not endorsed.
Note Dense 32B, strong Arabic; a common on-prem reference point
Note Large MoE, our upper reference for Arabic reasoning
Note Arabic-specialised 7B
Note 111B, strong RAG and multilingual
Note Saudi Arabic-first 9B
Note Hybrid architecture, good Arabic coverage (TII)
Note Compact multilingual baseline
Note Widely deployed 70B baseline
Note Efficient 24B for single-GPU appliances
Note Open-weight MoE, Apache-2.0
Note Jais 3 (Inception / G42, UAE) — Arabic-centric MoE