Mohamed bin Zayed University of Artificial Intelligence’s Institute of Foundation Models and Inception42, a G42 company, released Jais 3 on 15 October 2026. It is a family of six open, Arabic-first language models, ranging from 900 million to 375 billion parameters, together with a speech recognition model called Jais Speech. They were built in collaboration with Cerebras Systems.
Parameters are a rough measure of a model’s size. The developers said the smallest model is small enough to run on a phone. All six are built on K2 Horizon, the institute’s open foundation models. The developers said the models can take in up to 512,000 tokens of text at once, though the smallest model’s Hugging Face page lists 128,000. Jais Speech transcribes Modern Standard Arabic and dialects spoken across 19 Arab countries, keeps English or French words in their original script, and can identify the dialect.
The models can be downloaded from Hugging Face under the Apache 2.0 licence, which allows commercial use. The developers said they are releasing the set of 24 Arabic benchmarks they used, and that their largest model leads open models on agentic tasks. Those results are their own.
Why it matters: open Arabic models let Gulf universities, companies and government teams build tools in Arabic and run them on their own infrastructure, rather than relying only on systems designed mainly for English.
What to watch: independent tests of the developers’ results, and which services in the region start using the models.