APMIC's ACE-3 Tops AIEC Open-Source Language Model Evaluation
APMIC's Traditional Chinese language model ACE-3 ranked first in AIEC's latest evaluation of domestic and international open-source language models, scoring full marks on two of three Taiwan-focused indicators. APMIC said it fine-tuned ACE-1 with public Taiwanese corpora to strengthen Traditional Chinese and local-context understanding. The company also announced plans for PrivTuner, a self-service enterprise fine-tuning tool with more than 10 pretrained models, while registration is open ahead of planned trials in early 2027 and a formal release in the second quarter.
The AIEC result highlights a Taiwan-focused model and APMIC's enterprise fine-tuning roadmap, though the reported benchmark and product capabilities lack independent evaluation details.


台灣新創 APMIC(Accelerate Private Machine Intelligence Company)旗下繁體中文語言模型 ACE-3 ,在數發部 AI 產品與系統評測中心(AIEC)最新公布的國內外開源語言模型評測中名列第一。
APMIC 以 ACE-1 為基礎模型進行微調訓練,並在訓練過程運用數發部公開語料資源,強化模型對繁體中文與台灣在地語境的理解能力,打造出 ACE-3。
在 AIEC 涵蓋台灣在地語言與文化的 3 項評測指標中,ACE-3 於「台灣價值觀」與「高中學測國文」2 項指標獲 100% 滿分,「高中學測社會」也獲得 94.6% 優異成績,3 項指標合計達 294.6% 分,名列全球受測語言模型總排名第一。這凸顯 APMIC 透過適切的資料訓練與模型最佳化,既有模型也能持續強化在特定語言、知識領域及應用情境下的表現。