honey90/TenOS-Ko-28B

VISIONConcurrency Cost:2Model Size:27BQuant:FP8Ctx Length:32kTool Calling:SupportedPublished:Apr 26, 2026License:apache-2.0Architecture:Transformer Open Weights Cold

TenOS-Ko-28B is a 27 billion parameter multimodal language model developed by TenAI, based on the Qwen3.5 architecture and fine-tuned from Darwin-28B-KR. This model is specifically optimized for Korean language understanding, reasoning, and generation, excelling in areas like Korean culture, history, and legal knowledge. It supports multimodal inputs (image/video) and is designed to perform strongly on K-AI leaderboard evaluation items such as KMMLU-Pro and CLIcK.

Loading preview...

TenOS-Ko-28B: Korean-Optimized Multimodal LLM

TenOS-Ko-28B is a 27 billion parameter multimodal language model developed by TenAI, built upon the Qwen3.5 architecture and fine-tuned from the Darwin-28B-KR base model. Its primary focus is on enhancing Korean language capabilities, including understanding, reasoning, and generation, particularly for K-AI leaderboard evaluation metrics like KMMLU-Pro, CLIcK, MuSR(Ko), and Com2-main(ko).

Key Capabilities

  • Superior Korean Language Proficiency: Excels in Korean expression, cultural, historical, and legal knowledge.
  • Multimodal Support: Capable of processing image and video inputs.
  • K-AI Leaderboard Optimization: Specifically fine-tuned using domain-specific SFT for strong performance on key Korean AI benchmarks.
  • English Reasoning & Code-Switching: Demonstrates strong capabilities in English reasoning and English-Korean code-switching.

Good For

  • Korean-centric Q&A and Dialogue: Ideal for general conversations and specific inquiries related to Korean topics.
  • K-AI Benchmark Tasks: Performing well on K-AI leaderboard items requiring advanced Korean reasoning.
  • Multimodal Analysis: Analyzing images/videos and providing Korean descriptions.
  • Content Creation: Generating, summarizing, and creating various forms of Korean text.