I don't really understand the concept of Focal Length in OpenGL
https://en.wikipedia.org/wiki/Focal_length

It seems that this is related to the focus of the image and not the matrix or edges of the projection.

I also tested using the same settings for clip start and cilp end and apparently the origin point is in the expected position:

Perhaps the edges of the camera have to match the smallest side idk

Not sure how this is applicable to openGL but in context of cameras this is really just field of view. I think this report is not titled correctly. It may look like this is FL issue while it is camera placement issue. Steps I wrote down are destilled from last post in T43039.

Still not sure if this would be a bug, but it does look like a bug.

This change causes the projection matrix of the camera and the viewport to have the same field of view:

diff --git a/source/blender/blenkernel/BKE_camera.h b/source/blender/blenkernel/BKE_camera.h
index 812f5d520d7..2bcb055acec 100644
--- a/source/blender/blenkernel/BKE_camera.h
+++ b/source/blender/blenkernel/BKE_camera.h
@@ -88,7 +88,7 @@ typedef struct CameraParams {
 
 /* values for CameraParams.zoom, need to be taken into account for some operations */
 #define CAMERA_PARAM_ZOOM_INIT_CAMOB 1.0f
-#define CAMERA_PARAM_ZOOM_INIT_PERSP 2.0f
+#define CAMERA_PARAM_ZOOM_INIT_PERSP 1.0f
 
 void BKE_camera_params_init(CameraParams *params);
 void BKE_camera_params_from_object(CameraParams *params, const struct Object *camera);

It would also be necessary to change other default values such as the viewport depth of field and perhaps the initial camera's bounds.

That seems to resolve issue I outlined, so I would send that for review regardless. Still I would wait for response if this is actually issue originally reported maybe.