Xiaomi Launches MiMo-7B Reasoning Model Outperforming Competitors

Xiaomi has taken the lead in AI innovation with the release of its first open-source model specifically for reasoning tasks, named MiMo-7B. The model has achieved remarkable results in AIME (arithmetic and mathematical reasoning) and LiveCodeBench programming evaluations, surpassing leading competitors like OpenAI’s o1-mini and Alibaba’s Qwen-32B. MiMo-7B’s advancements stem from pre-training on vast amounts of reasoning data followed by reinforcement learning using innovative algorithms such as Test Difficulty Driven Reward. This model exhibits exceptional potential in reinforcement learning among similar 7B models, leading to a highly competitive edge.