AI & ML interests
Generative AI, Arabic NLP
Recent Activity
Inception42
Inception42, a G42 company based in Abu Dhabi, UAE, is AI-native from the ground up, transforming world-leading research into live, sovereign deployments at the core of governments and enterprises.
Our multilingual models are built for the languages, histories and cultures that frontier models weren't designed to serve, providing equitable access to AI.
Our vision and mission: A world where visionary thinking is a reality. Where every leader and organization has the certainty and confidence to solve their most critical challenges, and the freedom to think bigger and shape what's next.
Jais-3
Jais-3 is the third generation of the Jais family of Arabic-centric bilingual (Arabic–English) large language models. It spans dense models from 0.9B to 32B parameters, plus two Mixture-of-Experts (MoE) models for frontier-scale capability at a fraction of the active compute.
Try our Jais-3 in the Jais web app.
| Model | Uno variant | Architecture |
|---|---|---|
| Jais-3-0.9B | Jais-3-0.9B-Uno | Dense |
| Jais-3-3.7B | Jais-3-3.7B-Uno | Dense |
| Jais-3-7B | Jais-3-7B-Uno | Dense |
| Jais-3-32B | Jais-3-32B-Uno | Dense |
| Jais-3-MoVA-36B-A4B | — | Mixture of Experts (36B total, 4B active) |
| Jais-3-375B-A23B | — | Mixture of Experts (375B total, 23B active) |
Quantized and alternative formats (GGUF for llama.cpp/Ollama, MLX for Apple silicon) are available in the IFM/Jais-3 collection. Evaluation results, usage instructions and licensing details are on each model card.
Previous Generations
Earlier Jais releases remain available on Hugging Face:
| Generation | Description | Link |
|---|---|---|
| Jais-2 | The 2nd generation of the Jais family (Jais-2-8B-Chat, Jais-2-70B-Chat) | Jais-2 Family collection |
| Jais Family | Pre-trained and chat models from 256M to 30B, plus Llama-adapted models up to 70B | Jais Family collection |
| Jais | The original Jais-13B and Jais-30B releases | Jais Former Releases collection |
Copyright Inception42.
spaces 5
Multilingual Leaderboards 🌍
Generative Evaluation for Global South
AraGen Leaderboard
Generative Tasks Evaluation of Arabic LLMs
3C3H HeatMap
Generate heatmap visualizations of model evaluation metrics
X Risks Leaderboard
Study the Extreme threats of Frontier Large Language Models