ProASIC3, IGLOO nano, and iCE40 are all good, but the $2-3 ones just can't compete on I/O capacity with the Altera MAX II 240 and its 80 I/O in TQFP-100.
I requested a quote on a GOWIN part (GW1NR series) a while ago for a project that needed a smallish PLD and some SDRAM. The GW1NR series looked good since it integrates a 64Mbit SDRAM in the same package. Unfortunately for GOWIN it was just cheaper to go with the Altera MAX II and a separate SDRAM chip, so that's what I did.
I tried out the AGM software today, after much difficulty making a Baidu account in order to download it. After using the software, I see why AGM is not marketing their FPGAs in the US. In order to compile a design for the AG256, you must basically compile it for a Cyclone IV in Quartus, then the AGM software reads in the compiled netlist and fits it onto the much smaller AG256. (!) As promised by AGM, I was able to easily port a design from the Altera EPM240 to the AG256, but their synthesis approach is sketchy, to say the least. Nonetheless I ordered 10 AG256s and am excited to try them out in an existing design. Of particular interest to me are the current consumption and switching characteristics of the AGM chip in comparison to the Altera EPM240 and Lattice MachXO256.