DXwand Launches Prism v1.0 to Advance Arabic AI Accuracy

DXwand has introduced Prism v1.0, a small language model (SLM) designed to enhance how businesses interact with information in English and Arabic. Developed for accuracy and efficiency, Prism aims to set a new standard in AI-powered question-answering, particularly for Arabic-language applications.

Prism is available in two versions: a 1-billion parameter model, which is lightweight and cost-efficient, and a more advanced 3-billion parameter model, offering greater accuracy for complex queries. Both models are fine-tuned on LLaMA 3.2, significantly improving the benchmark for Arabic-language AI from 20% to 70%.

One of Prism’s key advantages is its minimal hardware requirements. It reduces GPU usage by over 85% compared to conventional AI models, allowing it to run on lower-cost hardware such as T4 and RTX GPUs. This efficiency makes high-performance AI more accessible to businesses without requiring extensive infrastructure.

The model’s design prioritises real-world applications, offering accurate, structured insights across multiple industries. Its multilingual capabilities ensure that users receive high-quality responses in both English and Arabic, a crucial factor for businesses operating in bilingual environments. DXwand has also integrated Prism with its ORXTRA Auto RAG Tuner, enabling businesses to refine the model’s outputs for their specific needs.

Ahmed Mahmoud, CEO of DXwand, described Prism as a step towards more practical AI adoption.

“We aim to bridge the gap between information and understanding, ensuring businesses have access to reliable and cost-efficient AI solutions. Prism reflects our vision for structured and actionable insights, making AI more accessible and effective for all.”

With a focus on efficiency, accuracy, and accessibility, Prism is set to enhance AI-driven knowledge discovery for organisations requiring reliable multilingual solutions.