ThakiCloud/Qwen3.8-27B-Human-KO-Enterprise-v0.1

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-v0.1 is a 27 billion parameter Qwen3.8-based model developed by ThakiCloud, specifically fine-tuned as an enterprise consultation agent for Korean language tasks. It excels at interpreting company policies and customer requests to select appropriate actions from a predefined set, demonstrating a 97.2% accuracy on enterprise operation benchmarks. This model is optimized for structured decision-making in customer service scenarios, outputting actions in JSON format.

Loading preview...

ThakiCloud/Qwen3.8-27B-Human-KO-Enterprise-v0.1 Overview

This model is a specialized checkpoint of ThakiCloud's Korean language model, ThakiCloud/Qwen3.8-27B-Human-KO, fine-tuned for enterprise consultation agent roles. It is designed to process company policies (regulations and exceptions) alongside customer requests and select a single, policy-compliant action from options like answering, re-asking, confirming, calling a tool, connecting to an agent, or refusing, outputting the choice as JSON.

Key Capabilities & Performance

  • High Accuracy in Enterprise Operations: Achieves 97.2% accuracy on the ThakiCloud/EnterpriseOps-KO-Policy benchmark, representing a 37.2% reduction in errors compared to the base model. This includes strong performance in refusing requests (97.1%), calling tools (80.5%), and handling exceptions (96.6% for not triggering exceptions, 99.4% for no exceptions).
  • Robust Safety & General Abilities: Maintains strong performance in general language understanding, with 92.7% on MMLU (English) and 68.0% on KMMLU (Korean). It also demonstrates high coding proficiency (94.5% on HumanEval) and effectively rejects all harmful requests.
  • Structured Output: Designed to output actions in a {"action": ...} JSON format, making it suitable for integration into automated systems.

Use Cases & Limitations

  • Ideal for: Automating customer service interactions, policy enforcement, and structured decision-making within enterprise environments, particularly for Korean language support.
  • Usage: Requires prompt_contract.txt instructions to be prepended to system or user messages, followed by policies and customer requests.
  • Limitations: Benchmarks are based on synthetic policies and requests; users should validate with their specific company policies and tools. The model selects actions but does not embed tool lists; actual tool call arguments must be handled by the service's schema.

Finetunes
1 models