I did some measurements on 3 of the best 2mm^2 LEDs, the dedomed XPG2 (new style), close sliced SST20, and white flat 2mm^2. The white flat is from Mouser and the item has a range of possible bins so I don’t know the exact bin. I measured the output by measuring the lux directly above the LED (80cm away) and assuming the light emitting surface is a Lambertian surface, which is a good approximation. See here some more information about this method. I use a Tondaj LX-1010B lux meter. The lumen numbers are approximately close to others’ measurements, but assume the usual 10-20% possible error.
I then measured the beam intensity of each LED in an EE X6 head powered by a constant current power supply. dedomed XPG2: 95.2kcd at 4.5A, sliced SST20: 92kcd at 4A, white flat: 111kcd at 6A. I calculated the luminance by dividing by the reflector area of 620mm^2 and assuming 90% reflector and glass efficiency. Then I scaled the output graphs according to this information and the luminance vs current is displayed below.
The white flat and dedomed XPG2 curves have close to the same relationship in the output and luminance graphs, which is expected since their dies have very similar areas. But the sliced SST20 luminance curve is high compared to its position in the output graph. I think this is explained by a non-uniform luminance across the SST20 die. The SST20 has a close slice, but obviously some silicone remains on the die, as pictured here. Near the edges of the die light can reflect off the silicone/air interface and exit the die to the side. This light doesn’t contribute to the light coming directly from the die; it bounces to the side and a higher fraction of it gets absorbed or emits to the side where it isn’t useful. You can see by looking at the lit die that the edges are less bright than the center, pictured below.
The other LED dies are much more uniform in brightness over their surfaces. I think this is the reason for the SST20 having a lower output than the other LEDs. But the measured luminance is high, even higher than the other LEDs for most of the curve. This is because measuring the luminance by measuring the beam center intensity from a reflector only measures the LED die-center luminance.
Conclusion:
These 3 LEDs perform relatively close to one another at a given current, but in real world use there are big differences. The dedomed XPG2 still performs well, but the high forward voltage means the current can be low, especially in a mulit-emitter light. The SST20 improves upon the XPG2 in this aspect with its lower voltage, but dedoming difficulties hold it back. The domeless white flat with its very low voltage and higher max output is the clear winner in some applications (multi-emitter), but driving a single emitter optimally might be more difficult.