Dexmal has unveiled a comprehensive embodied AI product lineup including the DM0.5 universal foundation model, the Apex universal robot body, and three enabling platforms: DexOS operating system, MaaS service platform, and DFOL 2.0 reinforcement learning framework. The strategy follows a three-stage rocket model: first stage, a generalized foundation model; second stage, developer tools including post-training, MaaS, and OS; third stage, systematic scenario solution capabilities. The DM0.5 foundation model, with 4 billion parameters trained on 150,000 hours of multi-source data, achieves 50ms inference latency, 25% improvement in inference efficiency, and 42% success rate on RoboChallenge Table30 V2 benchmark ranking first with 60.1 total score. On LIBERO evaluation, it achieves 99.1% comprehensive performance. Compared to DM0, zero-shot navigation success improved 31%, few-shot improved 45%, and fine-tuning success improved 20%. Fine-tuning costs dropped 60%, enabling expert-level deployment on a single consumer RTX 4090 GPU within 18 hours. DFOL 2.0 reduces real robot training data requirements by 60% and overall training costs by 40%. DexOS is the industry first embodied general-purpose operating system. The Apex universal robot features modular construction with hot-swappable end effectors in under 60 seconds, 3 kg single-arm payload, and 30-second hot-swap battery