We introduce the Aya Expanse model family, a new generation of 8B and 32B\nparameter multilingual language models, aiming to address the critical\nchallenge of developing highly performant multilingual models that match or\nsurpass the capabilities of monolingual models. By leveraging several years of\nresearch at Cohere For AI and Cohere, including advancements in data arbitrage,\nmultilingual preference training, and model merging, Aya Expanse sets a new\nstate-of-the-art in multilingual performance. Our evaluations on the\nArena-Hard-Auto dataset, translated into 23 languages, demonstrate that Aya\nExpanse 8B and 32B outperform leading open-weight models in their respective\nparameter classes, including Gemma 2, Qwen 2.5, and Llama 3.1, achieving up to\na 76.6% win-rate. Notably, Aya Expanse 32B outperforms Llama 3.1 70B, a model\nwith twice as many parameters, achieving a 54.0% win-rate. In this short\ntechnical report, we present extended evaluation results for the Aya Expanse\nmodel family and release their open-weights, together with a new multilingual\nevaluation dataset m-ArenaHard.\n