InternLM
State-of-the-Art Multilingual Open-Source Foundation Models with 1M Token Context and Advanced Reasoning.
The world's most advanced bilingual Arabic-centric LLM for sovereign AI.
Jais, developed by Inception (a Core42/G42 company) in collaboration with MBZUAI and Cerebras, represents a paradigm shift in regional AI sovereignty. Architecturally, it is a transformer-based decoder-only model trained on the Condor Galaxy supercomputer, utilizing a massive dataset of 395 billion tokens—including 116 billion Arabic tokens and 279 billion English tokens. Unlike Western-centric models that treat Arabic as a secondary language through limited tokenization, Jais employs a specialized tokenizer that captures the morphological complexity and nuances of Modern Standard Arabic (MSA) and regional dialects. By 2026, Jais has established itself as the foundational layer for MENA-region enterprise applications, offering superior performance in translation, summarization, and cultural reasoning compared to GPT-4 in Arabic contexts. Its integration into the Core42 Cloud ecosystem allows for high-throughput inference with regional data residency, making it the preferred choice for government, legal, and financial sectors requiring localized intelligence and strict data compliance.
A custom tokenizer built to handle the rich morphology of the Arabic language more efficiently than standard BPE tokenizers used by GPT.
State-of-the-Art Multilingual Open-Source Foundation Models with 1M Token Context and Advanced Reasoning.
Advanced AI reasoning with Constitutional safety for enterprise-scale cognitive tasks.
The definitive open-source framework for training and deploying massive-scale autoregressive language models.
The industry-standard LLM for high-throughput, cost-efficient natural language processing.
Verified feedback from the global deployment network.
Post queries, share implementation strategies, and help other users.
Utilizes Attention with Linear Biases to allow the model to extrapolate to longer sequence lengths than seen during training.
Trained on the world's largest AI supercomputer, ensuring massive parallelization and stability.
Incorporates Swish-Gated Linear Units to improve gradient flow and model convergence during training.
Specific variants (Jais-Chat) are fine-tuned on diverse instruction datasets to follow complex bilingual commands.
Safety filters and RLHF tuned specifically for MENA region cultural values and sensitivities.
The training set includes Modern Standard Arabic (MSA) as well as various regional dialects (Gulf, Levantine, Egyptian).
Providing citizens with 24/7 support in natural Arabic dialect.
Registry Updated:2/7/2026
Summarizing thousands of pages of Arabic legal text for quick review.
Analyzing social media and news for MENA stock market trends.