Ari Innovation Company announced that the platform (Spectrum!AI), the company’s AI and data solutions suite, achieved first place globally in the benchmark developed by Snowflake (Snowflake) in cooperation with the University of Oxford.
The platform’s artificial intelligence system recorded the highest score in the document attribution index across all participating systems, outperforming systems with open weight models (Open–Weight!System) and systems based on special programming interfaces (Proprietary!API–Based!Systems), thanks to its superior ability to accurately identify and cite correct source documents. The platform also outperformed its closest competitor by nearly 20 percentage points, and became the only open weight model system in the benchmark to exceed human performance, beating the human baseline by 5.5 percentage points.
The platform is designed to enable organizations to extract knowledge from their data by integrating high-accuracy OCR, information retrieval, independent intelligent inference, and reference citation verification. Unlike traditional systems that rely on prior knowledge of models, the platform reads and understands actual organization documents before generating answers, ensuring that outputs are based on verifiable sources.
The platform relies on its own OCR and independent thinking technologies powered by an 8-bit Coin model (Qwen3.6 35B–A3B8bit. bit). The open weight model version of the platform uses no more than 3 billion active parameters per token (Token). Despite its compact architecture, the system was able to outperform leading commercial models such as GPT-5 (GPT-5) and Claude Sonnet 4.5)Claude!Sonnet4.5) and Gemini 3 Pro (Gemini3Pro), which are widely believed to be much larger models.
The system can be deployed entirely within an enterprise infrastructure, with processing capacity exceeding 60 codes per second on a Mac (Mac) is just one, giving organizations a high-performance, low-cost option with complete preservation of their data within their own environment.
Commenting on this achievement, Alaa Saeed, founder and CEO of Ari Innovation, said:
“Data sovereignty, privacy and control are top considerations for customers when adopting AI solutions. Thanks to the creativity and excellence of the R&D and Quantitative Development teams at ARRI Innovation, customers no longer have to choose between these important requirements and high performance; they can have both.”
Ari Innovation also provides the platform via a managed API service (Managed!API), using platform-specific OCR and independent thinking technologies powered by the Gemini 3.5 Flash model (Gemini3.5Flash) This version achieved an accuracy of 91.8%, exceeding human performance by 9.6 percentage points. It also outperformed leading models from Google companies (Google(and Open AI)OpenAI)and anthropic(Anthropic(and perplexity)Perplexity(and Ali Baba)Alibaba) and others, including Gemini 3.5 Flash (Gemini3.5Flash) and GPT-5 (GPT-5) and Claude Sonnet 4.5)Claude!Sonnet4.5) and Gemini 3.1 Pro (Gemini3.1Pro).
For his part, Dr. Ali N. said: Salman, scientist and researcher in the R&D department at Ari Innovation Company and creator of the system that was evaluated:
“Our goal was not only to be the best in the region, but we designed the system to be among the best in the world, and the results have proven that. We are very proud that this innovation coming from Bahrain is able to compete globally.” Dr. Ali is a distinguished researcher with more than 20 international scientific publications, and specializes in privacy-preserving artificial intelligence and autonomous retrieval techniques (RAG), and recognizing facial expressions.
This achievement reflects Ari Innovation’s focus on developing AI systems that are not only accurate, but also auditable, easy to deploy, and practical suitability for enterprise environments.
In this context, Hussein Al-Musaed, Head of Quantitative Development Department at Ari Innovation, said:
“For regulated industries and organizations handling sensitive data, publishing via open-weight models offers a significant advantage. Data remains within the enterprise environment, costs remain predictable, and the entire system can be audited and customized to suit the structure, complexity, and documentation level of each client.”
