ThakiCloud/Qwen3.8-27B-Human-KO-Enterprise-v0.2

VISIONPricing:Input $1.6 / Cached $0.15 / Output $12Concurrent Unit Cost:2Model Size:27BQuant:FP8Context Size:32kTool Calling:SupportedPublished:Sep 24, 2026License:apache-2.0Architecture:Transformer Open Weights Featherless Exclusive Cold

ThakiCloud/Qwen3.8-27B-Human-KO-Enterprise-v0.2 is a 27 billion parameter Korean language model developed by ThakiCloud, fine-tuned for enterprise customer service applications. It excels at interpreting company policies and customer requests to select appropriate actions like answering, asking follow-up questions, calling tools, or escalating to a human agent, outputting decisions in JSON format. This model demonstrates high accuracy in policy adherence and shows strong performance in Korean and English knowledge, as well as coding tasks, making it suitable for automated customer support systems requiring precise policy-driven responses.

Loading preview...

Overview

ThakiCloud/Qwen3.8-27B-Human-KO-Enterprise-v0.2 is a 27 billion parameter Korean language model, specifically fine-tuned by ThakiCloud to act as an enterprise customer service agent. It processes company policies (including regulations and exceptions) alongside customer requests to determine the most appropriate action. The model outputs its decision as a single JSON object, choosing from actions such as answering, asking clarifying questions, confirming requests, calling tools, connecting to a human agent, or refusing a request.

Key Capabilities

  • Policy-Driven Decision Making: Accurately interprets company policies and customer requests to select the correct action.
  • High Accuracy: Achieves a 97.6% accuracy rate in enterprise policy scenarios, reducing errors by 44.7% compared to its base model.
  • Robust General Abilities: Demonstrates strong performance in English knowledge (MMLU 95.7), Korean knowledge (KMMLU 68.0), and coding (HumanEval 95.6).
  • Safety: Effectively rejects harmful requests, with 25/25 harmful requests refused.
  • JSON Output: Provides structured JSON responses for seamless integration into automated systems.

Good For

  • Automated Customer Service: Ideal for building intelligent customer support agents that adhere strictly to company policies.
  • Policy Enforcement: Useful for systems requiring precise interpretation and application of complex rules and exceptions.
  • Tool Calling: Can identify when external tools are needed and signal for their invocation, though actual tool arguments must be provided by the service.
  • Korean Enterprise Applications: Optimized for Korean language contexts in business and customer interaction scenarios.

Finetunes
1 models