So I’ve been playing around with the computer-use API, and so far it’s really cool and interesting… but my prices are absurd!
A simple command like “open firefox and go to github.com” costs $0.37, making me wonder if I’m implementing the API incorrectly.
Specifically, in regards to the screenshots, and how to optimize that somehow?
Small update - I scaled the screenshot resolutions down, which definitely helped, but costs went from $0.37 on average to like $0.25. It’s something to do with screenshot input, and whether that can be optimized or not.
Presumably Manus is doing something similar with Sonnet. My first thought when seeing their huge widescreen with all the agents going crazy was: “How much are they paying Anthropic for all these calls?!”