Use kimi 2.6 for planning and a cheap model (preferably local) for execution, and then kimi once again for reviewing it. Then finally I review the code. Saves a lot on tokens.
replyVery interesting, thanks for sharing. I am testing it with Pi in Zed and it seems pretty good.
reply