The Beijing Academy of Artificial Intelligence (BAAI) has unveiled the outcomes of the 'full-factor' verification for AI training, a task accomplished by FlagOS as part of the community intelligence program. Utilizing its comprehensive GPU computing card, the MTT S5000, which integrates AI training and inference capabilities, Moore Threads has effectively tailored the FlagOS training full-factor software stack. This adaptation facilitated the de novo (from scratch) training verification of the Qwen3-0.6B language model, utilizing 1T Tokens. Throughout this training process, the system maintained uninterrupted and stable performance for more than 14,000 steps, with the average relative error of the Loss curve kept within a tight 0.82% margin. When evaluated against standard downstream tasks, the system demonstrated a performance enhancement of 1.65 percentage points over the industry benchmark baseline set by NVIDIA.