The Scikit-learn 1.9 release brings significant improvements to existing machine learning models. Easily observable enhancements include richer HTML displays in notebooks, now showing fitted attributes and ColumnTransformer output feature names. A new, experimental callback mechanism has been introduced, allowing for progress bars and advanced monitoring during model training. This callback system is designed for flexible tracking of progress, even in parallel computing environments. Initially, callbacks are available for logistic regression,SearchCV objects, Pipelines, and StandardScaler. The release also focuses on improved statistics and numerics, enhancing the reliability of scikit-learn routines across diverse inputs and modeling choices. Tree-based models now have native support for missing values and monotonic constraints. Linear models benefit from float32 support in logistic regression and improved stability in RidgeCV/ClassifierCV. Scikit-learn now returns sparse arrays instead of sparse matrices, aligning with evolving SciPy practices. GPU support is expanding, with logistic regression, Poisson regression, and specific metrics gaining GPU acceleration. While user experience with GPU backends is still developing, it offers a prime area for contributor involvement. The scikit-learn project thrives on contributions from volunteers and financial sponsors.