国泰海通｜计算机：MiniMax发布全模态AI“全家桶”，M2登顶全球开源模型

Core Viewpoint - MiniMax, a Shanghai-based AI unicorn, has launched a comprehensive multimodal model suite called "全家桶," marking a significant breakthrough in China's AI technology landscape, particularly in multimodal capabilities [1][2]. Group 1: Product Launch and Performance - MiniMax's multimodal suite includes four major models: the text model M2, video generation model Hailuo 2.3, speech model Speech 2.6, and music model Music 2.0 [2][4]. - The text model M2 has achieved a remarkable position in the global rankings, being the first Chinese open-source model to enter the top tier of the Artificial Analysis (AA) leaderboard, with 10 billion active parameters and a total of 230 billion parameters [2][3]. Group 2: Cost Efficiency and Market Impact - M2 has set a new benchmark in model efficiency and cost control, with a reasoning cost as low as $0.53 per million tokens, which is only 8% of the cost of Claude 4.5 Sonnet, while achieving nearly double the reasoning speed [3]. - Following its release, M2's API call volume surged, ranking fourth globally and first among domestic models within just five days, demonstrating its strong market performance and potential for commercial application [3]. Group 3: Technical Innovations - The multimodal product matrix emphasizes generating quality and stability, with Hailuo 2.3 capable of producing 10-second native 1080p videos, and Speech 2.6 optimized for voice agent scenarios with a response time of 250 milliseconds [4]. - MiniMax's commitment to using a complete attention mechanism, despite industry trends favoring simplified versions, reflects its dedication to high-quality model performance in complex reasoning scenarios [4].