Spectrum, the Bahraini platform, clinches the lead in open-weight document models: 94% accuracy exempts Gulf banks from exporting their data

The Spectrum AI platform, developed in Manama, Bahrain, has achieved a continuous technical accomplishment by maintaining the global #1 position in the category of open-weight AI model systems for thirty consecutive weeks, according to the international evaluation standard (MADQA) that measures models' ability to understand, analyze and accurately extract data from documents. The latest version of the platform, developed by Ari for Innovation, a subsidiary of the portfolio of Bahrain Properties Holding, recorded a 6.7-point increase in accuracy to reach 94.4 percent, up from its previous result of 87.7 percent, thereby moving to third place globally among all systems evaluated, including closed-cloud models.

This engineering progress rests on the integration of optical character recognition (OCR) with retrieval-augmented generation (RAG), supported by intelligent agents designed to process dense enterprise documents such as contracts, financial records and regulatory reports. According to the evaluation results, the platform recorded an accuracy of 97.2 percent on the F1 metric used to measure the correctness of attributing answers to their original sources.These results demonstrate that open-weight models can achieve retrieval performance comparable to closed APIs while keeping processing entirely on self-hosted servers.

Commenting on the evaluation, Alaa Abdulkhaleq Saeed, the company's founder and chief executive officer, said that leading the global ranking for thirty weeks confirms the capability to build sovereign software hosted locally by institutions, ensuring the confidentiality of their digital assets. Dr. Saber Rahman, a senior AI scientist in the company's research and development department and holder of a PhD in theoretical physics from the Massachusetts Institute of Technology, explained that systematic experiments and system engineering have shown that open-weight models can match the efficiency of external APIs, allowing organizations to examine their sensitive files internally without sending them to cloud service providers.

This achievement moves the discussion on technology localisation in the Gulf, Egypt and the Arab Levant from theoretical ambition to concrete feasibility calculations. Regional banks and government agencies face strict regulatory constraints that prohibit sharing client documents and transaction records with external cloud platforms operating under foreign jurisdictions.Owning an open-weight model that reaches accuracy close to 95 percent implies redirecting budgets from cumulative cloud-usage fees toward investment in local computing servers.This shift also requires IT departments in the region to reshape their workforce skills, focusing on managing enhanced retrieval pipelines and running models within private data centres rather than relying solely on pre-packaged API access keys.

Placing a Bahraini model at the top of a specialised international ranking provides practical evidence that developing complex solutions is no longer the exclusive domain of traditional tech hubs. With governance and banking-compliance requirements expanding across regional markets, the ability to automate contract auditing and extract financial indicators from within an institution’s digital walls becomes a core pillar of operational security, paving the way for reduced reliance on external infrastructure in favour of a more independent and secure regional ecosystem.