Instead of Jupyter Notebook: marimo

Why marimo is better

• Running a cell reruns every cell that depends on it, so deleted or out-of-order cells cannot leave stale results. marimo cites a study where over a third of Jupyter notebooks on GitHub failed to reproduce.
• marimo notebooks are pure Python rather than JSON with embedded outputs, so reviews show real code diffs.
• A marimo notebook runs with python as a script and can be served as an interactive app, replacing papermill and Streamlit for many cases.
• Sliders and tables are reactive without callback code, and cells can query DataFrames and databases in SQL.

When to keep Jupyter Notebook

Keep Jupyter for kernels other than Python, such as R or Julia, for teaching material and platforms built around .ipynb, and for hosted services like JupyterHub setups that only serve Jupyter. Exploratory work that relies on redefining the same variable across many cells also fits Jupyter better.

Switching

Run marimo convert notebook.ipynb to get a .py notebook. The main adjustment is that a variable can be defined in only one cell, so repeated df = ... cells need merging or renaming. Expensive cells can be set not to autorun. JupyterHub can launch marimo through an extension.

Watch out

marimo supports Python only. The one-definition-per-variable rule takes getting used to. The ecosystem of extensions and hosted services is far smaller than Jupyter's, and many tutorials and courses still ship as .ipynb.

More newer, better dataframe libraries and notebooks

• pandas →, use Polars