The Singaporean government has embarked on an ambitious project to develop its own large language models (LLMs), aiming to achieve “sovereign AI” optimized for Southeast Asia’s multilingual and multicultural environment. This strategic move is intended to secure regional autonomy amidst the global AI race.
Driving the National Multimodal LLM Programme (NMLP)
The Infocomm Media Development Authority (IMDA) of Singapore, in collaboration with the Agency for Science, Technology and Research (A*STAR) and AI Singapore (AISG), is operating the ‘National Multimodal LLM Programme (NMLP)’. The National Research Foundation (NRF) is investing S70million(approximatelyUS52 million) into this two-year initiative, which is aligned with Singapore’s ‘National AI Strategy 2.0 and Research, Innovation and Enterprise (RIE) 2025 plan’.
The project is designed with three main objectives:
- Cultivating AI talent
- Enhancing industrial productivity
- Establishing a trustworthy governance environment
The initiative plans to provide high-performance computing resources to local researchers, encourage industries to adopt multilingual customer response systems and autonomous decision-making solutions, and build a trust-based AI usage environment by investigating LLM functionalities and risks.
MERaLiON and SEA-LION: Southeast Asia-Specific LLMs
The NMLP is driven by two core LLM models:
- MERaLiON (Multimodal Empathic Reasoning and Learning in One Network): Led by A*STAR’s Institute for Infocomm Research (I²R), this model focuses on recognizing and understanding code-switched conversations common in Singapore and Southeast Asia. It possesses multimodal processing capabilities that integrate text, speech, and scene information, and enhances contextual understanding by comprehending spoken language. Its six key functions include speech recognition/translation, speech summarization, speech Q&A, scene recognition, emotion and mood detection, and local dialect interpretation. It is expected to be widely applied in customer support, insight generation, and automated decision-making for both public and private sectors.
- SEA-LION (Southeast Asian Languages in One Network): Developed by AI Singapore, the national AI R&D promotion agency, this LLM is specialized in learning major regional languages such such as Thai, Vietnamese, and Indonesian. It aims to address the imbalance of linguistic representation and promote inclusive AI development by offering higher contextual relevance and response quality compared to Western or Chinese-centric models. Currently, the developed SEA-LION models are operating across Southeast Asia and are reported to outperform comparable open-source models.
Industry Collaboration and Sustainable Ecosystem Building
Industry participation is robust. A*STAR, in conjunction with IMDA, has formed the ‘MERaLiON AI Consortium’, with 13 organizations as initial members, including DBS Bank, Grab, Microsoft, SPH Media, and the National Supercomputing Centre (NSCC). They are jointly developing solutions for multilingual customer service, sentiment and health information analysis, and autonomous decision-making. The consortium aims to reduce development and integration costs by consolidating common needs and rapidly improve model capabilities through data and expertise sharing.
AI Singapore has also opened up external collaboration for expanding SEA-LION’s functionalities and enhancing its quality. The plan is to establish various partnerships based on open-sourced results to maintain a collaborative development framework. The Singaporean government anticipates that this project will simultaneously secure both regional relevance and strategic autonomy in AI technology.
