Xiaomi releases its first inference model MiMo-7B and announces full open source

PANews reported on April 30 that according to Xiaomi's official account "Xiaomi MiMo", Xiaomi opens its first big model MiMo-7B specially designed for "Reasoning" today, surpassing OpenAI o1-mini and Alibaba Qwen-32B in AIME mathematical reasoning and LiveCodeBench programming evaluation. MiMo improves its inference ability through large-scale inference data pre-training and reinforcement learning post-training, proposes innovative algorithms such as Test Difficulty Driven Reward, and leads in reinforcement learning potential among similar 7B models. The entire series has been open sourced to Hugging Face , and relevant technical reports have been released simultaneously.