Inception42 and Microsoft launch Seraj, an Arabic-first enterprise AI model
Seraj is built on GPT-4.1 and delivered through Core42's Compass platform, aimed at government, legal and financial workloads in Arabic.
Inception42 has launched Seraj, an enterprise AI model developed jointly with Microsoft to improve Arabic-language performance in government and business applications. The company announced the model on Thursday. It is available through Compass, Core42's sovereign AI platform.
Seraj is built on GPT-4.1 and takes its name from the Arabic word for light, or illumination. The model is intended for organisations that need precise Arabic output without losing capability in English or other languages.
The trade-off the model is meant to settle
Ashish Koshy, chief executive of Inception42, said organisations have been forced to choose between global AI capability and meaningful Arabic performance. Seraj removes that choice, he said, and is built so that organisations can deploy Arabic AI at scale with confidence.
Koshy described the launch as a turning point for the region's AI stack, and pointed to government services, legal analysis, customer engagement and knowledge management as the workloads the model was designed around.
Where frontier models fall short
Many leading models perform strongly in English but show meaningful gaps in Arabic linguistic precision, dialect comprehension, cultural context and safety handling. Those gaps have held back deployment in sectors that depend on Arabic fluency, trust and contextual understanding.
Inception42 addressed the problem through targeted mid-training on curated, high-quality Arabic datasets. The material covered linguistics, cultural knowledge, safety scenarios and domain-specific enterprise content. The company says the approach significantly improved Arabic performance while preserving the base model's reasoning and multilingual strengths.
What Seraj is built to do
The model is designed to handle document understanding, summarisation, translation, question answering, workflow automation, bilingual Arabic-English applications, retrieval-augmented generation and other knowledge-intensive tasks.
Target sectors include government, education, legal services, Islamic studies, media and financial services — fields where a mistranslated clause or a missed cultural reference carries operational and reputational cost.
Microsoft's position
Rima Semaan, director of AI and enterprise solutions at Microsoft UAE, said the collaboration with Inception42 reflects a shared commitment to expanding AI's real-world impact across the region. AI delivers the most when it can understand and engage people in the languages they use every day, she said.
Semaan said Seraj combines frontier model capabilities with linguistic and cultural intelligence that is relevant to the region, and supports the UAE's push to accelerate responsible AI adoption across governments, enterprises and critical industries.
Demand for Arabic-capable AI is rising across complex enterprise environments, particularly where precision, trust and contextual understanding are critical. More than 1 billion people use AI chatbots weekly, and experts warn of growing risks as adoption widens.
Delivery through a sovereign platform is the part worth watching. Governments and regulated industries in the region have been slow to commit to models they cannot govern locally, and a model that ships through state-controlled infrastructure gives procurement teams a path that does not require exporting sensitive workloads. Whether Arabic-first tuning becomes a standard requirement in enterprise tenders, rather than a differentiator, will depend on how quickly rival models close the language gap.
