Gemma Sea Lion V4 27B It

Gemma Sea Lion V4 27B It is a text-only instruct-tuned decoder model developed by AI Singapore's SEACrowd and AI Products Pillar, built on the Gemma 3 architecture as part of the SEA-LION family aimed at Southeast Asian languages. The IT release served as the post-training base for the companion vision-language variant Gemma-SEA-LION-v4-27B-VL, which extended the same Gemma 3 foundation with image-text capabilities across Burmese, English, Indonesian, Khmer, Lao, Malay, Mandarin, Tagalog, Tamil, Thai, and Vietnamese. The model is released under the Gemma Terms of Use and is distributed as open weights through the aisingapore Hugging Face organization, making it accessible for regional research, fine-tuning, and deployment scenarios that need strong representation of Southeast Asian languages alongside English.

The model's practical fit centers on multilingual natural language tasks for Southeast Asia, where the v4 line was designed to strengthen instruction following and conversational quality across the region's major languages. Because the IT variant shares Gemma 3's large context window and underlying decoder design, it is well suited for long-form document understanding, retrieval-augmented workflows, and structured generation tasks such as function calling when paired with the appropriate serving stack. Teams building Southeast Asia-focused assistants, content tools, or domain-specific chatbots benefit from a regionally tuned open-weight base, while developers needing multimodal reasoning can step up to the VL sibling for image-grounded question answering and document comprehension without leaving the same model family.