Use the most powerful model you can afford
The cheaper model costs more once you factor in your time.
A lot of AI advice now focuses on model efficiency. And since the recent price changes announcements from OpenAI, everybody is praising the lowest-tier, Luna.
Use Luna for most tasks. Upgrade to Terra for serious work. Save Sol for the “going to Mars” problems.
That approach makes sense when you are using the API at scale and paying for every token but matters much less when you are working through a subscription, which you should be.
Yesterday, I was debugging a file upload that kept failing. Terra spent six or seven turns changing the code, telling me the issue was fixed, and asking me to try again. No luck.
20 minutes in, I switched to Sol and asked it to inspect the problem and advise me.
In one turn, it identified the actual cause: a setting with the file hosting provider. The problem was never in the code.
At current API rates, Terra is half the price of Sol and Luna is one-fifth. Nice. But half an hour of your time is much, much more expensive.
My rule is now simple: use the most powerful model you can comfortably afford, especially when building things.
Then you can optimize down once you know the task is easy… or not.
If you’re spending more time managing AI than benefiting from it, let’s talk.