aisingapore/Gemma-SEA-LION-v4-27B-VL
Gemma-SEA-LION-v4-27B-VL is an instruct-tuned vision-text model developed by SEACrowd and AI Products Pillar, AI Singapore, based on the Gemma 3 architecture. With a 128K context length, this 27 billion parameter model excels at image and text understanding, including document comprehension, visual Q&A, and image-grounded reasoning. It is specifically post-trained on approximately 540k samples across 11 Southeast Asian languages, making it highly effective for SEA-specific tasks.
Loading preview...
Overview
Gemma-SEA-LION-v4-27B-VL is an instruct-tuned vision-text model developed by SEACrowd and AI Products Pillar, AI Singapore. It is built upon the Gemma 3 architecture and features a large 128K context length, enabling robust image and text understanding capabilities. The model has undergone extensive post-training using instruction-image pair datasets in 11 Southeast Asian languages, including Burmese, English, Indonesian, Khmer, Lao, Malay, Mandarin, Tagalog, Tamil, Thai, and Vietnamese.
Key Capabilities
- Multilingual Vision-Text Understanding: Excels in understanding and processing visual and textual information across 11 SEA languages.
- Document Comprehension & Visual Q&A: Capable of interpreting documents and answering questions based on visual input.
- Image-Grounded Reasoning: Performs reasoning tasks that integrate information from images.
- Advanced Function Calling: Supports structured outputs for seamless integration into larger systems.
Performance & Use Cases
As of October 2025, Gemma-SEA-LION-v4-27B-VL demonstrates strong performance on Southeast Asian tasks, often comparable to larger closed models and outperforming other open models under 200 billion parameters. It is particularly well-suited for applications requiring deep cultural and visual understanding within the SEA region. The model was evaluated using VQA tasks (MARVL, CVQA, WorldCuisines) and image captioning (XM3600) with a focus on SEA examples. Users should note that while its vision-text capabilities are strong, its text-only performance is comparable to its base model, Gemma-SEA-LION-v4-27B-IT.