Library

lammps stable_30Sep2026

Versionstable_30Sep2026
Stars★ 3,079
Released2026-09-30

Public development project of the LAMMPS MD software package

Release notes

Below is a list of major changes since the last stable release 22 July 2025

This stable release adds about 7400 commits since the previous stable release. This includes many contributions from beyond the core LAMMPS developers, but also continues the internal refactoring of the code base to make it more consistent and reliable, to take more advantage of modern C++ features where they help to simplify the code, and to make it easier to modify, maintain, and contribute to.

This is the first stable release where the impact of AI coding agents can be noticed: there are almost twice as many commits as between the previous two stable releases. AI coding agents have been predominantly applied in the following areas:
- port styles to the KOKKOS and OPENMP packages
- add more tests to the internal testing tree
- fix many long-standing bugs reported at https://github.com/lammps/lammps/issues or by static code analysis or running instrumented code (for memory issues)
- implement some requested features for which a reference code existed (e.g. in a different open source MD code)
- perform a thorough code review and correct identified issues

## Highlights

- LAMMPS now requires a C++17 compatible compiler and CMake 3.20 or later; the KOKKOS package requires C++20. The traditional GNU make build is limited to packages that do not need extra build steps in the `lib` folder, and LAMMPS-GUI now lives in its own repository.
- The the number of tests in the unittest tree has about doubled. Force style tests now also check the KOKKOS package with the Serial, OpenMP, CUDA, HIP, and SYCL back ends and the GPU package, in double, mixed, and single precision. Unit test and regression test results are collected on the new [LAMMPS Testing Dashboard](https://lammps.github.io/lammps-test-results/). Many bugs that were found while writing these tests, and by static code analysis with Coverity Scan and memory checkers, have been fixed.
- The number of KOKKOS styles more than doubled, and the OPENMP package gained about 100 new styles. KOKKOS now supports single and mixed precision, and has had many fixes for moving data between host and device. Obsolete code was removed from the GPU package (host/device particle split) and the INTEL package (Xeon Phi offload).
- Six new packages: FENIX, GRANSURF, GRAPHICS, MBX, ML-RUNNER, and QMMM-XTB and many updates and additions to existing packages.
- Much improved visualization with the new GRAPHICS package. The `dump image` and `dump movie` commands moved to the new GRAPHICS package and gained transparency, depth cueing, depth of field, outlines, specular and metallic highlights, gamma control, and background gradients. Many more kinds of objects can be rendered: dynamic bonds, regions, ellipsoids and superellipsoids, body particles, walls, indenters, surfaces, and atoms involved in Monte Carlo moves or reactions. Eight new `fix graphics/*` commands add arrows, chunk hulls, isosurfaces (with STL export), text and image labels and color bars, trajectory lines, periodic images, replica atoms, and more.
- New methods: a new FFT-based long-range solver (ESP), Gibbs ensemble Monte Carlo, BAOAB Langevin and ILVES constraint integrators, reversible SLLOD integration, BPM based peridynamics, granular surfaces and superellipsoid granular particles, the oxDNA3 model, and more.
- Styles are now stored in a process-global registry: loaded plugins persist across the `clear` command and are available to all LAMMPS instances in the same process.
- Refactored storage for variables; new `utils::sprintf()` function for more flexible format strings. Format strings for thermo, dump, and fix ave/time or ave/chunk output are now checked for consistency with the formatted data.
- Neighbor list binning can use a hash table, which avoids crashes and excessive memory use for very dilute or highly polydisperse systems.
- Fixed platform dependent results on ARM64 Linux and on platforms that enable fused multiply-add (FM…

Share this resource


Discovered 2026-10-02 Source GitHub Archive 2026-10 →