▲FFM-Mistral-Small-3.1-24B-Reasoning integrates Taiwanese values and optimizes local scenario model reasoning capabilities, supporting multimodal understanding, Hybrid Thinking mode, and vertical field applications.
ASUS Group's subsidiary, Taizhi Cloud, has launched a newly upgraded inference model, FFM-Mistral-Small-3.1-24B-Reasoning (Formosa Large Model FFM). This model integrates Taiwanese values and optimizes reasoning capabilities based on local scenarios, demonstrating superior reasoning and task understanding abilities. Supporting multimodal understanding and vertical application scenarios, the model meets the core needs of governments and enterprises in implementing autonomous, controllable, and trustworthy AI solutions, accelerating the deployment of Agentic AI and ushering in a new chapter in the development of sovereign AI.
Wu Han-chang, General Manager of Taiwan AI Cloud, stated, "This model supports flexible cloud and ground deployment options, effectively ensuring data sovereignty and sensitive information security. The multimodal understanding model can also be optimized to suit the local language and culture of each country, assisting governments in promoting the development of sovereign AI capabilities and empowering enterprises to apply localized, autonomous, and controllable AI models, accelerating the practical application of Agentic AI."
ASUS Associate President and Chief Technology Officer of Taizhiyun, Chen Zhongcheng, stated, "Taizhiyun continuously optimizes the latest open-source models globally, striving to enhance the localization and commercial application value of these models, providing stable and continuous commercial support services, eliminating model stagnation or upgrade challenges, strengthening enterprises' confidence and sustainability in adopting AI, and accelerating the deployment of enterprise Agentic AI applications."
Taizhi Cloud continues to develop a variety of open-source and reliable Formosa Large Models (FFM). Following the release of Llama3.3-FFM 70B and FFM-Embedding-v2 this year, the company has now launched FFM-Mistral-Small-3.1-24B-Reasoning, a lightweight model trained using model distillation and reinforcement learning techniques, effectively reducing hardware requirements and computational costs. In addition to supporting Traditional Chinese, it has been specifically enhanced for Taiwanese local knowledge, multi-turn logical reasoning, task decomposition, and data understanding. On TMMLU+ (Traditional Chinese Taiwanese local knowledge validation set), it achieves a score of 76.32%, significantly outperforming the native model in reasoning and task understanding evaluations.
FFM-Mistral-Small-3.1-24B-Reasoning is FFM's first model to support Hybrid Thinking, allowing users to flexibly initiate the reasoning process based on the scenario, adjusting performance and resource allocation to significantly enhance the model's general reasoning capabilities. Furthermore, this model can be specifically enhanced for government and industry scenarios to improve its problem-solving abilities.
As Agentic AI becomes a key trend in the next wave of artificial intelligence innovation, Taizhi Cloud will continue to deepen its model inference, multimodal capabilities, and experience in sovereign AI deployment. Combining the deployment flexibility of cloud-ground integration with the optimization capabilities of open-source models, it is committed to providing governments and industries with stable, reliable, and continuously evolving AI solutions.
▲FFM-Mistral-Small-3.1-24B-Reasoning possesses multimodal understanding of local Taiwanese scenes and reasoning capabilities. It can recognize images of Taiwanese tourist attractions and leverage reasoning abilities to deeply solve complex tasks.