Python backtesting libraries: what people use, and what is still maintained
Downloads tell you what people run. The commit history tells you whether anyone is still fixing it. Nobody publishes the two together, so this page does, refreshed every month.
Fetched 2026-09-03 from pypistats.org and the GitHub API. Free to cite with attribution and a link.
The finding
One library in this set has gone more than a year without a commit, and it accounts for 21% of the monthly downloads in the table. Backtrader was last touched 2024-08-19, which is 2.0 years ago, and still records 295,411 downloads a month.
Downloads and maintenance, September 2026
| Library | Downloads / month | Stars | Last commit | Licence |
|---|---|---|---|---|
| VectorBT | 519,570 | 8,664 | 2026-08-02 | NOASSERTION |
| Backtrader | 295,411 | 22,836 | 2024-08-19(2.0y ago) | GPL-3.0 |
| NautilusTrader | 292,184 | 25,462 | 2026-08-13 | LGPL-3.0 |
| Backtesting.py | 206,182 | 8,816 | 2026-08-05 | AGPL-3.0 |
| Freqtrade | 63,798 | 53,250 | 2026-08-13 | GPL-3.0 |
| Zipline Reloaded | 13,631 | 1,919 | 2026-01-06 | Apache-2.0 |
Downloads are the last 30 days from PyPI and include automated installs, so read them as a floor on real use. A commit older than a year is marked.
Stars and downloads measure different things
Freqtrade has the most stars in this set at 53,250, and it is not the most downloaded. VectorBT is, at 519,570 a month.
A star is one click, once, often from someone who never installed the project. A download repeats every time somebody builds an image or runs a pipeline. Neither is a quality measure, but if you are trying to guess how much company you will have when something breaks and you go looking for an answer, downloads are the better proxy.
How to read a stale repository
A frozen engine keeps whatever correctness it had. The risk is not that the maths rots, it is that the world moves: a data provider changes a column, a dependency ships a breaking release, a new Python version drops something the library relied on. None of that gets fixed.
So judge in two steps. First, does the engine compute what you need, which is what ourscored comparison and theexecuted benchmarks are for. Then ask whether anyone is still there when it breaks, which is what this table is for. A library can pass the first test and fail the second, and the download column shows that a great many people have not noticed.
Frequently asked questions
Which Python backtesting library is most used?
VectorBT, with 519,570 PyPI downloads in the last month. Downloads measure use rather than approval, and they include continuous integration and Docker builds, so treat them as a floor on real usage rather than a headcount.
Is backtrader still maintained?
Its last commit was 2024-08-19, which is 2.0 years ago as of September 2026, while it still records 295,411 downloads a month. The code still runs. What you do not get is fixes, and that matters most where an engine touches a moving target: broker APIs, data formats, and new Python releases.
Do GitHub stars tell you which library to use?
No, and this table is a clean demonstration. Freqtrade has the most stars at 53,250, and it is not the most downloaded. Stars measure attention at a moment, often from people who never install the thing. Downloads measure use, repeatedly. When the two disagree, downloads are the better guide.
Where do these numbers come from?
PyPI download counts from pypistats.org and repository signals from the GitHub API, fetched by a script in our public build and refreshed monthly. Figures on this page were last fetched 2026-09-03. Nothing here is estimated: a source that fails leaves the field blank rather than being filled with a guess.
Does an unmaintained library mean my backtest is wrong?
Not by itself. A frozen engine keeps whatever correctness it had. The risk is drift around it: your data provider changes a format, a dependency releases a breaking version, or a bug found by someone else never gets fixed. Judge the engine on what it computes, then judge the project on whether anyone is still there to fix it.
Related
Educational content only, not investment advice. Download and repository figures are third-party data reproduced as fetched, not measurements of software quality.