Powering the Next Wave of
AI in Southeast Asia
Explore our newest and most powerful SEA-LION models trained for Southeast Asia.
760k +
180k +
1 trillion
SEA-LION Model Family
Southeast Asia’s first family of open-source Large Language Models designed to solve these problems by deeply understanding Southeast Asia’s diverse contexts, languages, and cultures.
Latest Models
|
Name |
Version |
Parameters |
Quantised Versions |
Model Type |
Context length |
|---|---|---|---|---|---|
|
SEA-LION (Apertus) SEA-LION’s flagship model built on Swiss AI’s open-source models |
v4 (IT) |
N/A |
Instruct |
65K |
|
|
SEA-LION (Gemma) SEA-LION’s flagship model built on Google’s open-source models |
v4.5(IT) v4 (IT) v4 (VL) v4 (VL) |
N/A N/A N/A |
Instruct, Reasoning, Multimodal Instruct, Vision Instruct, Vision Instruct, Vision |
128K |
|
|
SEA-LION (Qwen) SEA-LION’s flagship model built on Qwen’s open-source models |
v4.5(IT) v4.5 v4 (IT) v4 (VL) v4 (VL) |
N/A N/A 8-bit, 8-bit (OpenVINO), 4-bit, 4-bit (OpenVINO) N/A N/A |
Reasoning, Agentic Drafter Instruct Instruct, Vision Instruct, Vision |
262K – 32K 256K 256K |
|
|
SEA-LION (Llama) Vision-language model for image + text comprehension |
v3.5(R) v3.5(R) |
Reasoning Reasoning |
128k 128k |
Ancillary Tools
|
Name |
Capabilities |
Details |
|
A collection of safety-focused LLMs built upon the SEA-LION family, these models are are specifically fine-tuned to detect, moderate, and handle content according to Southeast Asian cultural norms and safety standards |
Classifies prompts, response and prompt-response pairs into safe or unsafe |
|
|
Lightweight model for embeddings and retrieval |
Text classification and retrieval |
|
Updated as of 26 May 2026 – V4 Generation