Verbosity and number of steps for a simple task

[amaynez]

I use Cursor, Claude Code, ZenCoder and Kiro. Since even if I am paying I reach the hard limits very often, I wanted to give Kilo Code a try. I installed it and set it up with Kimi K2. I found that it consumes many more tokens than any other Coding LLM and it seems to repeat itself in what seems like an infinite loop before finding a trivial solution. For a very simple task I ended up using around $2 of my credits. Is it because of the model? Has anyone seen this happening? I tried both the debugging and the code agents.