Jais 3 understands spoken and written Arabic across dialects, produces Arabic text, and carries out practical work - from transcribing conversations to building documents and presentations. The fleet pairs six open language models, from 900 million to 375 billion parameters, with Jais Speech, a speech recognition model for Arabic and its dialects. All are built on K2 Horizon, IFM’s fully open-source foundation model fleet released in September 2026.
The models are trained on Modern Standard Arabic and the regional dialects - from the Emirati and Gulf dialects to North African Arabic – as well as the cultural knowledge the language carries for the more than 400 million people who speak it.
“Arabic AI needs to understand the language in its everyday context, which varies by geography and across generations,” said Haonan Li, senior research scientist at IFM. “For example, we need to account for the many nuances in spoken Arabic, including when speakers slip into English or French mid-sentence and back again. We built Jais 3 for Arabic as it’s really used.”
“Jais has shown what the UAE’s long-term investment in open AI research can deliver for the Arabic-speaking world. With Jais 3, we are extending that foundation into speech and agents at a time when the UAE is putting AI to work across government and society — reflecting MBZUAI’s commitment to the UAE National AI Strategy 2031 and to building the research capability and national talent that will carry it forward,” said Richard Morton, vice president and executive director of IFM.
IFM and Inception42 evaluated the models on an audited suite of 24 Arabic benchmarks, which 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. Jais Speech recorded the lowest average word and character error rates of any model compared, including commercial speech recognition services.
The advantage holds across sizes: the 3.7-billion-parameter model outperformed every competing open model of up to 5 billion parameters on Arabic and agentic tasks, and the 7-billion- and 32-billion-parameter models earned the highest average Arabic scores among open models at their sizes.
“Jais 3 makes Arabic AI more practical for business use. Stronger reasoning and instruction following, combined with tool use, provide a foundation for multi-step agentic workflows. With open models across a range of sizes, organizations can match these capabilities to their computing budgets and deployment needs,” said Dr. Larry Murray, SVP Applied Science at Inception42.
Jais Speech transcribes Modern Standard Arabic and dialects spoken across 19 Arab countries in real-world conditions. It follows speakers as they switch between Arabic and languages such as English or French, keeping those words in their original script, identifies the dialect being spoken and can add Arabic vowel marks to a transcript. Paired with the language models, Jais Speech lets Jais 3 take spoken instructions in Arabic or English and act on them.
“Jais has defined what an Arabic large language model can be since the first release, and Jais 3 is the clearest step up yet,” said Ashish Koshy, CEO of Inception42. “Building an Arabic LLM is not a translation exercise. It means training on the dialects people actually speak, the way they write, and the culture the language carries. That is the work we and IFM have done together across every generation of Jais, and it is why, by bringing together six new models, Jais 3 now leads on Arabic across every model size we have built.”
Jais 3 is the first adaptation of K2 Horizon to a new language. The six Jais 3 language models range from a 900-million-parameter model small enough to run on a phone to a 375-billion-parameter flagship. All share the same training recipe and agent capabilities, with a context window of 512,000 tokens, the longest of any Arabic-native model.
Developers can choose the size that fits their hardware and budget. At 7 billion parameters, the average score on IFM’s and Inception42’s Arabic evaluations rises from 56.4 for the K2 Horizon base model to 79.6 for Jais 3, showing that an open foundation model can be adapted to a new language and culture without starting from scratch.
Each model ships with five training checkpoints, letting researchers trace its development and build from any stage. Government entities and enterprises in the UAE and beyond can adapt the models and decide where to deploy them, in line with sovereign AI requirements.
