I have found it to be OK on sort of general purpose overview code stuff. At the moment it's too slow in llama.cpp on Mac for me to bother testing further.