For almost three decades, ReactOS has been the most ambitious impossible project in open source: a clean-room reimplementation of the Windows NT kernel that can run Windows applications without Microsoft's code. It has always been technically fascinating and practically borderline. Then, on July 5, 2026, a ReactOS nightly build ran Half-Life 2 in-game on real hardware.
The feat used a GeForce GTX 960 with NVIDIA's legacy 368.61 Windows driver — not an emulation layer, not WINE, not a VM. Just ReactOS, a GPU, and a game that shipped in 2004. According to Phoronix, this came roughly 30 days after a prior milestone: running the original Half-Life on actual hardware.
These aren't benchmarks. ReactOS's compatibility with real-world software is still incomplete, and the project's own developers are careful not to oversell it. But running a 3D game with a proprietary GPU driver is a meaningful signal about the state of ReactOS's Win32 subsystem and its ability to handle the kind of graphics and memory management that modern applications depend on.
The project started in 1996 as FreeWin95, pivoting to target Windows NT 4.0 after NT's design proved more principled and forward-compatible. Since then, progress has been glacial by necessity — every component of the OS must be written without referencing Microsoft's source code, and the team is largely volunteers working against one of the most complex software architectures ever built.
What makes the Half-Life 2 milestone notable is context. Valve's Source engine is not a trivial workload. It makes direct calls into Win32, talks to DirectX, and depends on driver behavior that most clean-room implementations struggle to replicate correctly. Getting Half-Life 2 running in-game means ReactOS's kernel, memory model, graphics subsystem, and driver stack all reached a threshold of simultaneous correctness that earlier versions couldn't sustain.
This also matters for the open source community beyond ReactOS itself. The project shares some infrastructure with WINE, and both efforts benefit from each other's research into how Windows internals actually behave under the hood. As ReactOS inches toward usability, the upstream benefit to WINE's compatibility with modern applications grows.
No one is predicting ReactOS will replace Windows on the desktop anytime soon. But for users who need to run legacy Windows software on hardware they control entirely — without telemetry, without license costs, without end-of-life risk — the gap between "interesting project" and "actually useful" is narrowing faster than it has in years. After 28 years, that's worth marking.