In general, or my personal one? In the former case, well, as accurate as you make it LinuxCNC suite supports 0,0001 mm accuracy, which can be further divided by using fraction of steps. Of course it's theoretical accuracy, pracital one need to be measured by real-life effect, and depends on the technique that your CNC machine is built, properties of materials used there, etc.

Another factor affecting accuracy, is the so-called "rectangle correctness" - i.e how close to ideal is the rectangle (or cube) made by your machine, in terms of angle accuracy. So it's not all about how well the machine reflects the step-motor move you want it to in linear move, the things like 3D-angles of every working part in your machine plays a role, too. Proper calibrating it for real-life results is a heck of a task, and if you're aiming for high accuracy, it should redone with every major change in temperature and humidity (unless you keep your CNC in something like server-room controlled environment).

You can skip most of those, if your machine have no more than something like 0,1 mm, accuracy (which, in real life effects, can create surprisingly non-matching parts...). The machine I'm using now - after many upgrades and design changes - have real-life accuracy of about 0,01 mm. Generally, whole things i (now) build for even better one, but the mechanic of the threaded axises is limiting here. One day, I will replace "classical" threaded axises I use, with much-more expensive ball-chain ones. For now, using backslash must be enough.
---

Now, just a small side-note - when comparing accuracy between different devices, be sure that you're looking at the test results from real-life things such machine have made. Many sellers (or even DIY'ers) post a theoretical accuracy, which is completely unrelated to reality (that's why you may read about 0,01 accuracy on 3D printer, while, in reality, you still have "stripes" of plastic, that you need to clean manually, and are visible clearly by naked eye, or felt to the touch).

/Estel