scalejade/qwen-sea-lion-v4-32b-it

TEXT GENERATIONPricing:Input $0.408 / Cached $0.0816 / Output $1.972Concurrent Unit Cost:2Model Size:32BQuant:FP8Context Size:32kTool Calling:SupportedPublished:Aug 18, 2026License:mitArchitecture:Transformer Open Weights Featherless Exclusive Cold

Qwen-SEA-LION-v4-32B-IT is a 32-billion parameter instruction-tuned large language model developed by AI Singapore, built upon Qwen3-32B. It is specialized for Southeast Asian languages, having been continuously pre-trained on approximately 100 billion tokens from the SEA-Pile v2 corpus across seven SEA languages. The model supports a 32,768-token context window and is designed for multilingual assistants, translation, RAG, summarization, and instruction-following tasks in the Southeast Asian region.

Loading preview...

Overview

Qwen-SEA-LION-v4-32B-IT is a 32-billion parameter instruction-tuned large language model developed by AI Singapore, based on the Qwen3-32B architecture. It has been specialized for Southeast Asian (SEA) languages through extensive continued pre-training and instruction fine-tuning. The model was pre-trained on approximately 100 billion tokens from the SEA-Pile v2 corpus, covering seven SEA languages: Burmese, Indonesian, Malay, Filipino, Tamil, Thai, and Vietnamese. Following this, it underwent post-training on about 8 million high-quality instruction pairs.

Key Capabilities

  • Multilingual Proficiency: Strong foundation in Burmese, English, Indonesian, Khmer, Lao, Malay, Mandarin, Tagalog, Tamil, Thai, and Vietnamese.
  • Extended Context Window: Supports a native context length of 32,768 tokens.
  • Instruction Following: Designed for general instruction-following tasks.
  • Qwen3 Thinking Mode: Inherits Qwen3's optional enable_thinking mode for enhanced reasoning.

Intended Use Cases

  • Multilingual Assistants: Building conversational agents for SEA languages.
  • Translation: Facilitating translation between English and various SEA languages.
  • Retrieval-Augmented Generation (RAG): Enhancing RAG systems with regional content.
  • Summarization and Classification: Performing text summarization and classification tasks.
  • Cultural Context Tasks: Handling tasks requiring an understanding of SEA cultural nuances.
  • Domain-Specific Fine-tuning: Serving as a robust base model for further specialization.

Limitations

Users should be aware that the model has not been safety-aligned and may hallucinate or produce factually incorrect content. It is not recommended for applications requiring hard safety guarantees or critical advice without additional safety fine-tuning and evaluation.

Finetunes
1 models