MiniMax announces expansion of m3 pro model and domestic computing power infrastructure

MiniMax, a leading artificial intelligence company, announced during its mid-year earnings call that the M3 Pro model is expected to have a parameter scale of approximately 3T. This enhancement aims to improve the model's generalization capabilities and intelligent limits through expanded reinforcement learning and long-term task training. MiniMax's founder and CEO, Yan Junjie, stated that the company will continue to advance the synergy between model intelligence, inference efficiency, and infrastructure capabilities, accelerating the development and release cycle of the M series and H series models. Additionally, MiniMax is working on adapting the M3 and H3 models to domestic chips, with a large-scale domestic computing power cluster set to go live soon and gradually take on real production traffic. According to IT Home's previous report, MiniMax released its 2026 fiscal year first half (January to June 2026) report, which showed a total revenue of $11.7 million (approximately 788 million RMB), representing a 283.1% year-over-year increase. The gross profit was $20.813 million (approximately 140 million RMB), a 464.8% year-over-year increase. The net profit attributable to shareholders was -$358 million (approximately -2.412 billion RMB), with a narrowing loss of 11%. The company also officially open-sourced the general-purpose video model MiniMax H3, which supports multi-modal context and 2K resolution.