Last updated:
SEA-LION
AI Singapore's SEA-LION — open models tuned for Southeast Asian languages and contexts.
Model Facts
Latest model: SEA-LION-Embedding
- Provider
- AI Singapore
- Released
- Aug 1, 2025
- Context window
- 128K tokens
- Weights
- Open
Full Model List
| Model | Released | Status | Context | Note |
|---|---|---|---|---|
| SEA-LION-Embedding | Mar 1, 2026 | current | – | embedding model suite; month approximate |
| SEA-LION v4 27B | Aug 1, 2025 | current | 128K tokens | 128K context, supports RAG and agents; month approximate |
- Official model page:
- sea-lion.ai
- Other sources:
- SEA-LION v4 27B (LLM Reference)
- AI Singapore (LLM Reference)
What it is
SEA-LION (Southeast Asian Languages in One Network) is AI Singapore's family of open models. The SEA-LION v4 line spans roughly 7B to 70B parameters, with the 27B instruction-tuned model offering a 128K-token context and support for retrieval-augmented generation and agents. It recognises 13 languages, including Javanese, Sundanese, Malay, Tamil, Thai and Vietnamese. A SEA-LION-Embedding suite followed in March 2026, and AI Singapore is working with Dell to bring SEA-LION to AI PCs and edge devices.
Where this information comes from
Details come from AI Singapore coverage and a model tracker; release months are approximate. Release dates, context windows and benchmark scores are vendor-reported unless stated, and harnesses differ between labs, so treat the numbers as indicative.
Notable traits
- Built for Southeast Asian languages and cultural context.
- Open models from a national programme.
- Small variants for edge devices.