A screenshot from the linked article titled "Reflection in C++26", showing reflection as one of the bullet points listed in the "Core Language" section
It's the capability of a program to "reflect" upon itself, I.E. to inspect and understand its own code.
As an example, In C# you can write a class...
public class MyClass
{
public void MyMethod()
{
...
}
}
...and you can create an instance of it, and use it, like this...
var myClass = new MyClass();
myClass.MyMethod();
Simple enough, nothing we haven't all seen before.
But you can do the same thing with reflection, as such...
var type = System.Reflection.Assembly.GetExecutingAssembly()
.GetType("MyClass");
var constructor = type.GetConstructor(Array.Empty<Type>());
var instance = constructor.Invoke(Array.Empty<Object>());
var method = type.GetMethod("MyMethod");
var delegate = method.CreateDelegate(typeof(Action), instance);
delegate.DynamicInvoke(Array.Empty<object>());
Obnoxious and verbose and tossing basically all type safety out the window, but it does enable some pretty crazy interesting things. Like self-discovery and dynamic loading of plugins, or self-configuration of apps. Also often useful when messing with generics. I could dig up some practical use-cases, if you're curious.
It's pretty cool when you use it right but it's also really easy to shoot yourself in the foot with, even by C++ standards. For example, in other languages (I'm coming from Java/C# which both have it) it lets you access private/protected fields and methods when you normally wouldn't be able to.
There's also a noticeable performance penalty over large lists because you're searching for the field with a string instead of directly accessing it.
For the times it is necessary (usually serialization-adjacent or dynamic filtering/sorting in a table) to use reflection, it's faster at runtime than converting an object to a dictionary/hashmap. However, 99% of time it's a bad call.
If you look at the proposal, this is specifically "static reflection", i.e. compile-time reflection. So it doesn't actually have any of the downsides you mention, as far as I can tell.
Not a C++ dev, but looking at Java, which has reflection: Reflection allows to inspect and modify runtime attributes of classes, interfaces, fields and methods. Even, when you don't know their names at compile time.
Basically take any object and just ask "what are your (even private) fields?" and then happily modify them, or call these methods, or...