Shopee Built Custom AI Model Named Compass
The e-commerce giant developed its own multilingual language model to optimize fraud detection and platform operations.
Updated on Sept. 24, 2026 in Artificial Intelligence

Live Poll
Do you believe companies building custom AI models will provide you with a better shopping experience?
Shopee has launched Compass, a multilingual large language model designed to handle complex e-commerce tasks across Southeast Asia, Taiwan, and Brazil. The platform built the model using the NVIDIA AI Factory to address limitations in general-purpose AI.
Why it matters
General-purpose models previously struggled with Southeast Asian e-commerce requirements, resulting in lost revenue and slower support. Compass replaces these legacy systems to improve operational efficiency and security for the company.
Compass-v3 is a mixture-of-experts model featuring 245 billion total parameters and 71 billion active parameters per token. It was trained on 12 trillion tokens of multilingual data and provides 50 times faster fraud detection.
The players
Shopee
Shopee is an international e-commerce platform that facilitates over $136 billion in annual gross merchandise value across eight markets.
NVIDIA
NVIDIA is a technology company that provides the AI Factory infrastructure used to train advanced large language models.
The details
Shopee utilized the NVIDIA AI Factory to build this custom architecture, which includes pretraining, instruction following, and agentic reinforcement learning. The system now manages the majority of the company's artificial intelligence traffic while reducing production costs by 90 percent.
Timeline
Monthly API token usage increased from 3 billion to 340 billion over the past eight months.
The Tech Race
The shift toward specialized, domain-specific models like Compass reflects a move away from reliance on general-purpose AI for complex enterprise tasks. Companies are increasingly building proprietary mixture-of-experts architectures to gain performance advantages over mass-market competitors.
Users will experience faster and more accurate fraud detection when shopping on the platform. The transition to a custom model also aims to improve support resolution times for international customers.
The takeaway
Specialized AI models often outperform larger, general-purpose systems by focusing on the specific linguistic and structural needs of a particular industry. Businesses can achieve significant cost savings and performance gains by investing in infrastructure tailored to their own internal data.
Further reading
Learn more about the latest innovations in Artificial Intelligence.
Source note: This article includes information reported by Frontier Enterprise.
Live Poll
Do you believe companies building custom AI models will provide you with a better shopping experience?



