中国独角兽开源M2比肩硅谷技术，成本仅8%

Core Insights - MiniMax, a Chinese AI unicorn, has launched and open-sourced its new text model, MiniMax-M2, which features 10 billion active parameters and a total of 230 billion parameters, achieving lower latency and cost while enhancing throughput efficiency [1][20][22] - The MiniMax-M2 model has ranked in the top five globally and first in open-source on the Artificial Analysis (AA) leaderboard, outperforming models from Google and Anthropic [1][20][21] - The model's cost-effectiveness is highlighted, with its comprehensive cost being less than 8% of Claude 4.5 Sonnet, aiming to disrupt the expensive "computing power game" [1][22][23] Performance and Features - MiniMax-M2 is designed for coding and agent tasks, showcasing strong end-to-end development capabilities and the ability to handle multiple code files [21][22] - The model excels in various benchmarks, ranking second globally in deep search capabilities and financial search benchmarks [21][22] - It supports up to 200,000 tokens, making it suitable for complex tasks and workflows [4][21] Market Reception - The model has received positive feedback from overseas developers and tech influencers, with notable endorsements from platforms like HuggingFace and LMarena [1][23] - MiniMax-M2's API service pricing is set at $0.3 per million input tokens and $1.2 per million output tokens, significantly lower than competitors [22][23] Strategic Implications - The launch of MiniMax-M2 reflects a shift in China's role in the global AI landscape, moving from "Made in China" to "Created in China" [26][27][28] - The model's development aligns with China's strategic goals of achieving technological self-sufficiency and fostering original innovation in AI [27][28] - MiniMax aims to attract global developers for further innovation and application of its model, enhancing China's influence in AI governance and standards [27][28]