◈
My Experience With Claude "Effort" Levels
Let’s be honest here, if you’re a professional developer in 2026, there’s an extremely high chance you’re using AI to some degree or another for coding assistance. I do it more than I’d like to admit. My favorite AI to code with is Claude. It’s both creative and skilled at programming, not to say it’s without its flaws. And I’m not just talking about ethics. I’ve noticed over the last few days that not only do different levels of effort from Claude AI give you varying levels of quality for more or fewer tokens, but the correlation between effort and quality is not what you may expect.
Let me cut to the chase: Low effort seems to be the best option both for using fewer tokens and getting a higher quality response faster. I’ve noticed Claude seems to make more assumptions when trying to be higher effort, where as low effort Claude asks about the information it doesn’t know more often than high effort. On top of this, higher effort takes longer to generate an answer and consumes more tokens. In medium and high effort, I often wait 10-15 minutes for a response only to find that I ran out of free tokens and the response was incomplete, and therefore I get no response. In high effort, often times it doesn’t even finish thinking before I run out of free tokens. Low effort, however, seems to be the sweet spot in all areas - better quality answers due to fewer assumptions and less overthinking, quicker responses due to less thinking, and fewer tokens consumed.
I wouldn’t go as far as to say higher effort Claude is a scam, but I do believe that it doesn’t work as intended, *from my experience. With that said, I encourage you to run your own testing to figure this out for yourself. I didn’t exactly do extensive testing; I’m just speaking from experience.
Addendum
While we’re on the subject of Claude, I’ve noticed that Fable 5 and Mythos 5, Claude’s newest premium commercial AI models, are currently unavailable. I looked into why, and the reason is actually very surprising: the U.S. government ordered Anthropic to revoke access to the model for national security reasons. Anthropic only has a vague hint as to why they were ordered to do this, but they believe it’s due to jailbreak concerns. With this in mind, Anthropic disagrees with the U.S. government’s decision, but is complying anyway. You can learn more here.