aisingapore/Qwen-SEA-LION-v4.5-27B-IT

VISIONConcurrent Unit Cost:2Model Size:27BQuant:FP8Context Size:32kTool Calling:SupportedPublished:May 14, 2026License:mitArchitecture:Transformer0.0K Open Weights Featherless Exclusive Cold

Qwen-SEA-LION-v4.5-27B-IT by AI Singapore is a 27-billion parameter instruction-tuned causal language model built on the Qwen3.6 dense architecture, featuring a hybrid Linear and Full Attention design. It boasts a 262K context window and strong reasoning capabilities, with specific post-training for multilingual and multicultural fluency across English and key Southeast Asian languages including Burmese, Indonesian, Filipino, Malay, Tamil, Thai, and Vietnamese. This model excels in agentic coding, unified vision-language tasks, and retains historical reasoning context for iterative development.

Loading preview...

Overview

Qwen-SEA-LION-v4.5-27B-IT is a 27-billion parameter Large Language Model developed by AI Singapore, specifically designed for the Southeast Asia (SEA) region. Built upon the Qwen3.6 dense architecture, it incorporates a hybrid Linear and Full Attention design and features a substantial 262K context window. The model underwent distillation from Qwen3.5-397B-A17B using an updated aisingapore/SEA-Instruct-2602 dataset, ensuring deep domain adaptation and strong multilingual and multicultural fluency.

Key Capabilities

  • Multilingual Proficiency: Fine-tuned for English and seven key Southeast Asian languages: Burmese, Indonesian, Filipino, Malay, Tamil, Thai, and Vietnamese.
  • Extended Context Window: Inherits a 262K context window from Qwen3.6, enabling robust reasoning over long inputs.
  • Advanced Reasoning: Designed as a highly capable reasoning model with configurable thinking modes and the ability to preserve historical reasoning context.
  • Agentic Coding: Offers high-precision handling for repository-level reasoning and frontend workflows.
  • Unified Vision-Language: Features early fusion training for strong performance across multimodal reasoning, coding, and visual tasks.
  • Tool Calling: Demonstrates capabilities in tool use, as shown in examples for searching HDB listings and calculating mortgages.

Good For

  • Applications requiring strong language understanding and generation in Southeast Asian languages.
  • Tasks benefiting from a large context window, such as complex document analysis or multi-turn conversations.
  • Development of agentic systems, particularly for coding and structured data interaction.
  • Multimodal applications that integrate vision and language processing.

Limitations

  • The model has not been aligned for safety, requiring developers to implement their own safety fine-tuning and security measures.
  • It may exhibit limitations common to LLMs, such as hallucination and occasional generation of irrelevant content.