On April 30, according to Xiaomi's public account "Xiaomi MiMo", Xiaomi today opened up the first large model MiMo-7B specifically created for "Reasoning", surpassing OpenAI o1-mini and Ali Qwen-32B in AIME mathematical reasoning and LiveCodeBench programming evaluation. MiMo improves reasoning capabilities through the linkage of large-scale reasoning data pre-training and post-reinforcement learning training, proposes innovative algorithms such as Test Difficult Driven Reward, and strengthens learning potential leadership among similar 7B models. The entire series has been open source to Hugging Face, and relevant technical reports have been released simultaneously.