OLMo 2 is the latest fully open language model launched by Ai2, available in both 7B and 13B parameter sizes, trained on a massive dataset of 5T tokens. These models perform comparably or even better than other fully open models of similar size and compete with open-weight models like Llama 3.1 on English academic benchmarks. The development of OLMo 2 emphasizes stability during model training, staged training interventions, cutting-edge post-training techniques, and an actionable evaluation framework. These technologies enable OLMo 2 to excel across multiple tasks, particularly in knowledge recall, common sense, and both general and mathematical reasoning.