JTS-AI/OpenJAI-v1.0-14B

TEXT GENERATIONConcurrency Cost:1Model Size:14BQuant:FP8Ctx Length:32kTool Calling:SupportedPublished:Oct 6, 2025License:apache-2.0Architecture:Transformer0.0K Open Weights Cold

OpenJAI-v1.0 is a 14 billion parameter open-source Large Language Model developed by Jasmine Technology Solution (JTS), built upon the Qwen3-14B architecture. It is specifically fine-tuned for high performance in both Thai and English, excelling in instruction following, long-context understanding up to 120,000 tokens, and reliable tool calling. This model improves upon its base and outperforms other leading open-source Thai models of comparable size across diverse benchmarks, making it ideal for applications requiring strong multilingual capabilities and advanced practical skills.

Loading preview...

OpenJAI-v1.0: A High-Performance Thai and English LLM

OpenJAI-v1.0 is a 14 billion parameter Large Language Model developed by Jasmine Technology Solution (JTS), leveraging the robust Qwen3-14B foundation. This model is meticulously fine-tuned to deliver state-of-the-art performance in both Thai and English, with a strong focus on practical applications.

Key Capabilities & Enhancements

  • Thai-Centric Excellence: Achieves top performance in Thai while maintaining strong English capabilities.
  • Enhanced Practical Skills: Excels in complex instruction following, long-context understanding, and reliable tool/function calling.
  • Long-Context Understanding: Supports robust performance on input contexts up to 120,000 tokens (with appropriate scaling techniques like YaRN for contexts exceeding its native 32,768-token window).
  • Tool Calling: Demonstrates strong capabilities for agentic use, compatible with frameworks like Qwen-Agent.
  • Performance Gains: Outperforms its Qwen3-14B base model and other leading open-source Thai models of similar size across various benchmarks, including instruction following (IFBench-EN/TH), multi-turn capability (MT-Bench-TH), long-context understanding (MRCR, LongBench-v2), and tool calling (BFCL-v3-EN/TH).
  • Knowledge Retention: Fine-tuning was conducted without significant degradation of the base model's core knowledge, avoiding catastrophic forgetting.

When to Use OpenJAI-v1.0

  • Applications requiring high-performance Thai and English language processing.
  • Scenarios demanding complex instruction following and multi-turn conversations.
  • Use cases involving very long documents or conversations (up to 120,000 tokens).
  • Developing agentic systems that require reliable tool and function calling.
  • Research and development within the Thai AI community due to its fully open-source nature.

Finetunes
1 models