A new Thai large language model named OpenJAI-v1.0-14B has been released by Jasmine Technology Solution (JTS), built upon the Qwen3-14B model. This 14-billion parameter model has been further trained to enhance instruction following, long-context understanding, and tool-calling capabilities, while aiming to avoid catastrophic forgetting of the base model's knowledge. The model's performance is detailed in a research paper on arXiv, showing it outperforms its base model in 7 out of 10 test sets, particularly excelling in Thai tool-calling tasks. AI
IMPACT This Thai-language model could accelerate adoption of LLMs for specific regional applications, particularly in tool-calling tasks.
RANK_REASON Release of a new open-source LLM with performance details and a research paper. [lever_c_demoted from research: ic=1 ai=1.0]
- Apache Software License 2.0
- arXiv
- BFCL-v3-TH
- GPT-4.1 nano
- JaiTTS
- Jasmine Technology Solution
- MMLU-ProX-lite-TH
- Model Context Protocol
- MT-Bench-EN
- Nokka
- OpenJAI-v1.0-14B
- Qwen3-14B
- Qwen-Agent
- vLLM
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →