Xiaomi Unveils MiMo-7B Model Surpassing Competitors in Reasoning Tasks

According to PANews, Xiaomi has announced the release of its first open-source model designed for reasoning, named MiMo-7B. The model has demonstrated superior performance in AIME mathematical reasoning and LiveCodeBench programming evaluations, outperforming OpenAI's o1-mini and Alibaba's Qwen-32B. MiMo-7B enhances reasoning capabilities through pre-training on large-scale reasoning data and subsequent reinforcement learning, introducing innovative algorithms such as Test Difficulty Driven Reward. It leads in reinforcement learning potential among similar 7B models. The entire series has been made available on Hugging Face, accompanied by a detailed technical report.