Pick The Right Model And Effort Level
From the course: Agentic AI For Founders & Senior Operators
Default to Sonnet 5.5 on high effort, especially on the Pro plan, and switch to Opus 5.5 on medium effort for complicated or strategic work. Opus tends to give better output because its agent loop runs longer, though a very high effort level can make it overthink. There's no set rule for when to switch, so find out by trial and error, and drop to a lower model if your usage starts running out.
Transcript
Okay, so this is the chat window that you should be very familiar with by now. You will see here on the bottom right, this is the model and effort selector. So when I click on that, I can switch the model and I can switch the effort level. So let's say from here, these are the three most recent options. Sonnet is typically quite okay, especially if you're on the Pro plan. I recommend that you use Sonnet and that you use a high effort level. The only real difference that you will find is the quality of output with Opus, which is the one that I use, tends to be better. And that's just because the agent loop tends to run for longer. You can experiment with it. Typically for strategic thinking, you may wish to use Opus. For daily ops type things where you're just pulling data or something, you could use Sonnet. I just use Opus for everything because I have the token budget. But if you don't, then go with Sonnet for your default. Switch up to Opus when you're doing something complicated. There's no real guidance on how to determine whether to switch a model or an effort level. It is something that you figure out using trial and error. Okay, so that's the piece on the model side. The key takeaway is use Sonnet 5.5, use high effort, otherwise switch to Opus 5.5 and use medium effort. You can always check your usage and confirm if you're starting to run out, then just switch to a lower model. Okay, it's really just to do with how long the agent loop runs. If you run an effort level of very high, in fact, you'll see that it tends to start overthinking. So which is why for Opus 5.5, I recommend using a medium effort level.