Xiaomi's Core Large Model team has proudly announced the open sourcing of its pioneering large-scale inference model, aptly named "Xiaomi MiMo." This model leverages both pre-training and post-training phases to significantly bolster its inference capabilities. In public evaluations conducted on mathematical reasoning (AIME 24-25) and code competitions (LiveCodeBench v5), MiMo, with a modest 7 billion parameters, impressively outperformed OpenAI's proprietary inference model o1-mini and Alibaba's substantially larger open-source model QwQ-32B-Preview. This accomplishment underscores Xiaomi's formidable innovative prowess in the realm of artificial intelligence.
