I’m really liking the experience of using Cline with OpenRouter. You get detailed token count and context size for every Task as it goes, along with the actual price in realtime. That combined with the ability to use virtually any model with a single key makes it easier to determine what work to give to what model, and if it’s worth it to you. Use the best models for the most important work, cheap/free ones for the rest, and do the rest yourself. That way you drive your model, not when the model decides enough is enough.