ThakiCloud/Qwen3.8-27B-Human-KO-Finance
ThakiCloud/Qwen3.8-27B-Human-KO-Finance is a 27 billion parameter model, based on the Qwen3.8 architecture, specifically fine-tuned for Korean financial securities compliance. Developed by ThakiCloud, this model is designed to observe mandatory disclosures and the suitability principle in client advisory conversations without explicit prompting. It excels in ensuring regulatory adherence in financial advice, particularly when used with its mandatory gate mechanism to achieve 100% disclosure coverage and zero factual errors.
Loading preview...
ThakiCloud/Qwen3.8-27B-Human-KO-Finance Overview
This 27 billion parameter model, developed by ThakiCloud, is a specialized version of the Qwen3.8 architecture, fine-tuned for Korean financial securities compliance. Its primary function is to automatically adhere to mandatory disclosure requirements and the suitability principle during client advisory interactions in the Korean financial sector.
Key Differentiators & Capabilities
- Mandatory Gate Mechanism: Crucially, the model's performance metrics are achieved only when deployed with its accompanying "gate" mechanism. This gate ensures 100% disclosure coverage and eliminates factual errors, which are not guaranteed by the weights alone.
- Compliance-Oriented: Designed to observe specific Korean regulatory guidelines, including the Korea Financial Investment Association's Standard Investment Recommendation Guidelines and Article 19 of the Financial Consumer Protection Act.
- High Suitability Behavior Pass Rate: Achieves an 81.8% suitability behavior pass rate, significantly outperforming its baseline (FIN2).
- Zero Factual Errors: When used with the gate, the model demonstrates zero factual errors on a 200-item evaluation set.
- Synthetic Data Training: The model was trained on synthetic data generated by ThakiCloud's own serving model.
Important Limitations
- Gate Dependency: The model's superior performance in disclosure and factual accuracy is strictly dependent on the use of the provided
gates/component. Deploying without it will result in missed disclosures. - Statistical Precision: Suitability measurements have a statistical spread due to the limited evaluation set size (n=33 judgeable items).
- Legal Review Required: Despite its capabilities, legal review is still necessary before deployment into an actual regulatory environment, as the gate catches shortfalls but does not replace legal oversight.