The Open-R1 initiative has unveiled a series of new open-source AI models, notably the OlympicCoder, which includes two variants: a 7 billion parameter (7B) model and a 32 billion parameter (32B) model. The 7B model has demonstrated superior performance on the International Olympiad in Informatics (IOI) benchmark, outperforming the Claude 3.7 Sonnet model. The 32B variant has excelled against all tested open-weight models, including those significantly larger, marking a notable advancement in competitive programming capabilities. Additionally, the OLMo 2 32B model has been introduced, which is the first fully open model to surpass both GPT-3.5 and GPT-4o mini on a variety of multi-skill benchmarks, while requiring a fraction of the training compute typically needed for such performance. All datasets, training recipes, and benchmarks for these models are being made publicly available, further promoting open-source AI development.