Mohamed bin Zayed University of Artificial Intelligence's Institute of Foundation Models and Inception42 have introduced Jais 3, a new fleet of Arabic-native artificial intelligence models for speech, writing, and agents. The release builds on the K2 Horizon foundation model family that IFM launched in September 2026. It pairs six open language models with Jais Speech, a speech recognition system designed for Arabic and its dialects.
Designed for Arabic as It Is Spoken
Jais 3 understands spoken and written Arabic across regional dialects, from Emirati and Gulf varieties to North African Arabic. The models are trained on Modern Standard Arabic and the cultural knowledge carried by more than 400 million Arabic speakers. They also account for code switching, including moments when speakers move between Arabic, English, and French in a single conversation.
A Family of Models Across Sizes
The Jais 3 fleet includes six open language models ranging from a 900-million-parameter model small enough to run on a phone to a 375-billion-parameter flagship. All models share the same training recipe and agent capabilities, with a context window of 512,000 tokens, which IFM describes as the longest of any Arabic-native model. Developers can choose the size that fits their hardware and budget needs.
Speech Recognition and Practical Deployment
Jais Speech transcribes Modern Standard Arabic and dialects spoken across 19 Arab countries in real-world conditions. It can follow speakers as they switch between Arabic and languages such as English or French, preserving those words in their original script. The system also identifies the dialect being spoken and can add Arabic vowel marks to a transcript.
Benchmark Results and Agentic Tasks
IFM and Inception42 evaluated the models on an audited suite of 24 Arabic benchmarks that they are releasing publicly alongside the models. The largest Jais 3 model leads open models on agentic tasks and matches the strongest on Arabic understanding, while Jais Speech recorded the lowest average word and character error rates among compared systems. At 7 billion parameters, the average Arabic evaluation score rises from 56.4 for the K2 Horizon base model to 79.6 for Jais 3.
Open Access and Sovereignty
Jais 3 is the first adaptation of K2 Horizon to a new language, demonstrating that an open foundation model can be customized for a specific language and culture without starting from scratch. Each model ships with five training checkpoints, allowing researchers to trace its development and build from any stage. Government entities and enterprises can adapt the models and decide where to deploy them, supporting sovereign AI requirements.
Strategic Context and Collaboration
Jais 3 was created by the Institute of Foundation Models at MBZUAI and Inception42, in collaboration with Cerebras Systems. Richard Morton, vice president and executive director of IFM, said the release extends Jais into speech and agents at a time when the UAE is putting AI to work across government and society. Dr. Larry Murray of Inception42 added that stronger reasoning, instruction following, and tool use provide a foundation for multi-step agentic workflows.
Executives from IFM and Inception42 framed Jais 3 as a step toward more practical Arabic AI for business and government. The release reflects the UAE's long-term investment in open AI research and its alignment with the UAE National AI Strategy 2031. Jais 3 and Jais Speech are available today on Hugging Face, with deployment support for vLLM and SGLang.