Someone on Threads asked me how I split my budget across these AI subscriptions. I already had a framework in my head. I wrote about the usage economics before; this post is the concrete conclusion on where the money should actually go.
Claude first. On the same plan, quota usage feels relatively durable to me, as long as it doesn’t get reset. If you’re going to subscribe to Claude, my suggestion is simple: go straight for the 100 USD plan. Not because of quota size, but because their biggest edge is Fable. Fable’s architectural strategy, intent alignment, and insight are still ahead of anything GPT has put out, as far as I’ve seen.
Then Codex. 20 USD is enough, no need to go higher. Luna Max inside it, used for browser automation and computer-use tasks, is the best cost-performance combo I’ve used. I occasionally open Sol for adversarial review, and 20 USD covers that plenty. If it’s not enough, I just bet on a Tibo reset. When it hits, it’s a great day again.
Postscript: I Chose to Bravely Downgrade
I used to feel so embarrassed about it. Everyone’s tokenmaxxing, showing off how many 200U plans they’re burning through. If I can’t use all of mine, does that mean I’m useless and my productivity is too low?
That mindset is actually really unhealthy. People who use a lot don’t need to judge how others use their tokens. Using less can also mean you’ve already got your routing sorted. It’s not that we aren’t trying or that our output is low.
Give a bit more time back to yourself. Tokens will always be there if you subscribe again, but life is only 30,000-some days, and time is the one asset you never get back.