小米首个推理大模型开源 7B参数规模能力超越OpenAI o1-mini

Core Insights - Xiaomi has launched its first reasoning-oriented large model "Xiaomi MiMo," which enhances reasoning capabilities through a combination of pre-training and post-training [1] - In public evaluation sets for mathematical reasoning (AIME24-25) and coding competitions (LiveCodeBench v5), MiMo outperformed OpenAI's closed-source reasoning model o1-mini and Alibaba's larger open-source model QwQ-32B-Preview, using only 7 billion parameters [1] Company Developments - The introduction of Xiaomi MiMo marks a significant advancement in the company's AI capabilities, focusing on reasoning tasks [1] - The model's performance indicates Xiaomi's competitive position in the AI landscape, particularly against established players like OpenAI and Alibaba [1] Industry Trends - The development of reasoning models is becoming increasingly important in the AI industry, with companies striving to enhance their models' capabilities [1] - The competition among AI firms is intensifying, as demonstrated by Xiaomi's ability to surpass larger models with fewer parameters [1]