Anserwise/AWAXIS-Think-28B

VISIONConcurrency Cost:2Model Size:27BQuant:FP8Ctx Length:32kTool Calling:SupportedPublished:Apr 24, 2026License:apache-2.0Architecture:Transformer0.0K Open Weights Cold

Anserwise/AWAXIS-Think-28B is a 27 billion parameter, multilingual causal language model developed by Anserwise, specifically optimized for Korean language reasoning and general-purpose inference. It features a Qwen3.5 hybrid attention architecture with a 262,144 token context length, derived from an evolutionary merge of Qwen3.6-27B and a Claude Opus-distilled reasoning model, further fine-tuned with a minimal Korean LoRA. This model excels in complex reasoning tasks and natural language understanding across Korean, English, Chinese, and Japanese.

Loading preview...

AWAXIS-Think-28B: Korean-Optimized Reasoning LLM

Anserwise/AWAXIS-Think-28B is a 27 billion parameter, multilingual large language model (LLM) primarily focused on enhancing reasoning capabilities, especially for the Korean language. It is built upon a unique evolutionary merge process using the Darwin V7 system, combining the strengths of two base models:

  • Qwen/Qwen3.6-27B: Providing a robust, general-purpose backbone.
  • rico03/Qwen3.6-27B-Claude-Opus-Reasoning-Distilled: Imbuing the model with advanced reasoning abilities distilled from Claude Opus.

This merged model, named Darwin-28B-Opus, then underwent a minimal Korean LoRA fine-tuning (r=8, attention-only) to significantly improve Korean language naturalness and contextual understanding while preserving its strong general and reasoning capabilities. The model utilizes a Qwen3.5 hybrid attention architecture (48 Linear Attention layers + 16 Full Attention layers) to efficiently handle a substantial 262,144 token context length.

Key Capabilities

  • Advanced Reasoning: Inherits strong inference capabilities from its Claude Opus-distilled parent model.
  • Korean Language Specialization: Enhanced naturalness and contextual understanding for Korean through targeted LoRA fine-tuning.
  • Multilingual Support: Proficient in Korean, English, Chinese, and Japanese.
  • Long Context Handling: Efficiently processes long inputs up to 262,144 tokens with its hybrid attention mechanism.

Good For

  • Korean Q&A, dialogue, and creative writing.
  • Complex, long-form reasoning and Chain-of-Thought applications.
  • Multilingual tasks involving Korean, English, Chinese, and Japanese.
  • General summarization, translation, and text analysis.

Finetunes
1 models