Government-Certified Trustworthiness
APMIC's Traditional Chinese AI model is among the first in Taiwan to pass the Trusted AI Language Model Assessment conducted by the AI Evaluation Center (AIEC), a government-accredited AI testing body. This ensures that government agencies and legal entities at all levels can adopt the model with confidence in its security, controllability, and regulatory compliance.
AIEC is the AI evaluation body established under the Administration for Digital Industries, Ministry of Digital Affairs, tasked with developing internationally aligned AI model testing methodologies and trust frameworks.
Trained on Traditional Chinese data, balancing performance with governance needs
ACE-2-27B is developed by APMIC’s elite AI team using the PrivModel fine-tuning and distillation solution. It is currently the first 27B-scale Traditional Chinese model purpose-built for enterprise applications. The model supports 128K long context, is fully FP4 quantized, and is fully compatible with the NVIDIA Blackwell architecture, offering comprehensive improvements in performance and inference speed.
The training data covers three major Traditional Chinese domains: professional knowledge, science and technology, and Taiwan humanities and social sciences, enabling the model to handle complex reasoning and long-text tasks effectively. It supports stable on-premise deployment while balancing inference performance, response consistency, and enterprise data governance requirements — helping organizations achieve reliable AI reasoning and knowledge integration.
Official document and report generation
Document review
Government document retrieval
Note: This application is powered by PrivAI. The interface shown is for demo purposes only. Actual implementation can be customized based on client requirements or integrated into existing enterprise systems.
ACE-2-27B
Model | Compute Architecture | Hardware Requirements | Recommended Hardware | Context window | Features |
|---|---|---|---|---|---|
ACE-2-27B | APMIC PrivModel | Recommended GPU VRAM above 48 GB | NVIDIA RTX PRO 6000 Blackwell Series *1,
NVIDIA DGX B200/B300 *1 | 128K Tokens | Optimized for MCP interaction |
Fully on-premises deployment |
Taiwan’s first 27B model |
FP4 quantization × Blackwell architecture: improved performance and faster response |
Trained on Traditional Chinese datasets, covering knowledge, science, and Taiwan humanities & social sciences |
Gemma-3-27B | Google fine-tuning and distillation technology | Recommended GPU VRAM above 48 GB | NVIDIA RTX PRO 6000 Blackwell Series *1,
NVIDIA DGX B200/B300 *1 | 128K Tokens | Fully on-premises deployment | Supports multiple languages and diverse tasks | Based on Google Gemini model technology |
Mistral-24B | Standard dense Transformer architecture | Recommended GPU VRAM above 48 GB | NVIDIA RTX PRO 6000 Blackwell Series *1,
NVIDIA DGX B200/B300 *1 | 32K Tokens (128K tokens for version 3.1) | Fully on-premises deployment |
Supports table data and basic document processing tasks |
ACE-1-24B
ACE-1-24B is a proprietary Traditional Chinese reasoning model developed by APMIC. Featuring a 65K long-context window, it is specifically optimized for enterprise applications. Trained on three major Traditional Chinese datasets—covering mathematical logic, daily reasoning, and tool instructions—it supports Chain of Thought (CoT) reasoning. ACE-1-24B enables efficient on-premise deployment, ensuring stable responses and seamless knowledge integration.
ACE-3B
ACE-1-3B is the first 3B parameter Traditional Chinese language model in Taiwan that can run on mobile devices. Via the Model Context Protocol MCP it connects to internal and external enterprise systems to enable natural language-driven information retrieval workflow operations and decision support, creating a new intelligent human-computer interaction interface. It also supports integration with external APIs, ERP systems, and IoT devices.
Sovereign AI
Native Linguistic & Cultural Mastery
Goes beyond generic models to deeply understand local laws, social norms, and specific semantic logic (e.g., Traditional Chinese and industry jargon).
Secure Integration & Data Privacy
Supports on-premise or private cloud deployment for total data sovereignty. Seamlessly connects with internal systems for rapid, secure deployment.
Domain-Specific SOP Fine-tuning
Tailors AI to manufacturing, healthcare, or finance by integrating internal SOPs and proprietary data through PrivModel technology.
Contextual Compliance & Knowledge Base
Ensures precision in government, legal, or PR sectors by aligning with local values and syntax, eliminating the biases of general-purpose AI.
%2020250916_zhiting_ace-1%E7%B6%B2%E9%A0%81.png)
Financial advisor model
Understands 6 financial domains
Covers basic financial planning, risk management and insurance planning, employee benefits and pension planning, investment planning, and tax and wealth transfer planning. Tailor-made for financial consultants, solving client needs in a single place.
Seamless integration with internal KM systems
Connects to company knowledge bases and process guidelines to stay current on product information, compliance policies, and sales scripts.
Fine-tuned with internal SOP and data
Uses PrivModel technology to feed company documents and SOP into the model, creating a company-specific AI.
Passed TABF financial license exams for stronger professionalism
In internal testing, it passed question banks from the Taiwan Academy of Banking and Finance, including compliance items such as Anti-Money Laundering, demonstrating professional and regulatory awareness beyond general-purpose models.
A thinking AI advisor with scenario-based interactive responses
The AI does not just answer. It analyzes step by step, based on the customer profile and questions, before giving advice, creating warmer financial communication.
Product Management & Logistics Model
Truly understands manufacturing
Integrates data and processes across production, logistics, R&D, and QA, focusing on the most critical pain points on the shop floor.
Full ERP and MES data integration
Supports most ERP connections to onboard internal data quickly and activate intelligent applications with minimal effort.
Fine-tuned with internal SOP and data
Using PrivModel technology to feed company documents and SOP into the model, creating a company-specific AI.
Proactive alerts and intelligent decision recommendations
Beyond knowledge base use, it analyzes anomaly root causes, proposes multiple solutions, and accelerates decision-making.
5 key AI applications to boost efficiency and decision quality
Build scenarios such as downtime alerts, intelligent scheduling, inventory analysis, BOM checks, and automatic 8D report generation.
%2020250916_zhiting_ace-1%E7%B6%B2%E9%A0%81.png)
FAQ
We’ve compiled the most frequently asked questions about implementing the ACE series models. Whether you’re in the early evaluation stage or ready to launch your project, the following answers will help clarify your direction and address any concerns.
The ACE series has been fine-tuned and distilled specifically for Traditional Chinese data and Taiwan-based use cases. It performs more accurately in Traditional Chinese reasoning and TMMLU+ benchmarks, delivering results that better align with local language needs and achieve higher scores.
ACE-1-24B is designed for tasks that require high accuracy and deep reasoning, such as financial analysis, legal review, and medical research. It supports MCP tool invocation capabilities, allowing flexible integration with external systems and data sources. It also features Agent capabilities, enabling it to autonomously plan and coordinate multi-step tasks—making it ideal for complex projects and enterprise-level applications.
Yes. The ACE series will continue to expand, offering different versions and industry-specific models optimized for varying hardware conditions and application scenarios, allowing enterprises to flexibly select the most suitable reasoning model.
Yes. ACE supports the MCP protocol and standard APIs, enabling seamless integration with ERP, CRM, knowledge base, and IoT systems.
You can contact the APMIC team through the “Contact Us” page to discuss adding new data or adjusting model directions. APMIC will provide expert recommendations and assist you in maintaining and optimizing the model to ensure long-term alignment with enterprise needs.
ACE-1-24B features a 65K context window, supporting multi-turn reasoning and dialogue. In enterprise scenarios, it maintains contextual coherence and delivers more reliable insights.
It integrates with ERP/MES systems for use in downtime prediction, intelligent scheduling, inventory analysis, BOM inspection, and automatic 8D report generation—specifically designed to enhance efficiency and decision-making in manufacturing environments.
The model has passed tests based on the Taiwan Academy of Banking and Finance (TABF) question sets and compliance standards, including Anti-Money Laundering (AML) requirements. It can also integrate with internal KM systems and operational workflows to ensure both professionalism and regulatory compliance.
Talk to Our Team
Ready to accelerate your AI capabilities? Our team is here to help you explore how APMIC’s PrivModel solution—purpose-built for high-efficiency, enterprise-grade AI—can support your business goals. Whether you're looking to customize a private LLM, simplify deployment, or unlock more value from your data, we’re ready to help you take the next step.