One of the reasons I liked the GM1020 versus the UT382 was the range. The UT382 maxes out at 20k lux, the GM1020 goes high enough to measure direct sunlight.
I hadn't really tested this feature much, but I gave it a workout yesterday. It did extremely poorly. The situation was a 10 lumen flashlight. I was interested in how the flashlight did when the battery was nearly dead and the output weak. So to maximize the sensitivity I set the flashlight on top of the luxmeter, at point blank. This produced a reading of around 25k lux and should let me measure the output down to the last few millilumens.
The runtime plot was disappointing. Note that there are no connecting lines between samples. Each measurement is a single dot. The Y axis is the measured lux and the X axis is the sample count.
The behavior above 20k lux is terrible. First, it is noisy. It jumps all over the place by about 10%. Second, it has no resolution. See the horizontal bands? The meter simply can't measure a value between the bands. Third, the transition to the lower range has a discontinuity and jumps from 20k to 19k.
These flaws don't mean it is a bad meter. You will never see 20k lux in your integration sphere. Even for candela measuring, it only happens with the brightest throwers at close range. (Use a 90% neutral density filter or put some more distance between the meter and the light.) Keep it under 20k lux and the readings are solid.