Anthropic's reasoning budget controls are changing how I prompt Claude

Liam Wilson · 0 likes

Yeah I do this constantly now. Low budget for boilerplate, max for anything touching concurrency bugs. Night and day difference in cost.

Jasmine Cruz · 0 likes

same here, our bill dropped like 30% just from this one change

Mariam Said · 0 likes

honestly i just leave it on high always, my time is worth more than the token cost lol

Midnight Rider · 0 likes

the tricky part is predicting which tasks NEED the budget before you run them. sometimes a 'simple' refactor hides a gnarly edge case

Gentle Noodle · 0 likes

this. ive been burned a few times assuming low effort was fine

Jonas Schneider · 0 likes

curious if anyone has benchmarked this against just using Opus for everything instead of toggling Sonnet's budget

Grace Parker · 0 likes

we built a tiny router that classifies task complexity first then picks budget automatically, works pretty well

Grace Parker · 0 likes

i feel like antrhopic shoud expose more granular telemetry on where the thinking tokens actually go, its kinda a black box rn

Ngozi Bello · 0 likes

lol my coworker set it to max for a one-line css fix and it thought for 45 seconds, we laughed so hard

Chloe Davis · 0 likes

genuinely one of the more useful knobs theyve shipped, makes cost scale with actual difficulty instead of flat rate