it's ok but not great compared to the commerical LLM's it's really not very good. it's benchmarks are clearly fake. even still we run it internally for a ton of workloads on our rack of gpu's