Xiaomi released the first inference model MiMo-7B and announced that it is fully open-source

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