Moore Threads Trains 236B MoE and Embodied AI Models on MTT S5000

During its semi-annual business briefing, Moore Threads addressed investor and partner concerns about operational readiness. Chairman and CEO Zhang Jianzhong stated that the company's overall supply chain and production capacity are fully aligned with its business development needs, ensuring timely delivery of AI computing solutions as demand surges.

Zhang also highlighted significant technical milestones. Using Moore Threads' self-developed MTT S5000 AI computing cluster, the company has trained a 236-billion-parameter Mixture-of-Experts (MoE) base large language model from scratch. It further completed the full training pipeline for frontier models such as RoboBrain2.5, an embodied intelligence brain model, and EvoPhys-World, a 5D world model, developed with the Beijing Academy of Artificial Intelligence (BAAI). These achievements demonstrate Moore Threads' capability to enable large-scale AI training and advance embodied intelligence applications, strengthening its position in the domestic AI accelerator market.