"This article could have been a video" (did a deep dive on how to maximize usage in gpt-5.6 on your Codex subscriptions)

got 5.6 sol is nerfed and it stops mid process and asks for confirmation or to continue. goal also doesn't work as well. getting pretty much same performance to 5.5

It's a nice video but I think it could have been an article instead

This setup has worked remarkably well for me, so i decided to share it. I get out way more of my limits than before.

Do you have any videos on these things but for us API-only users? At my work we do have access to adolescent, but only through API and fixed monthly budgets. Sol can *melt* those

Most of this applies there as well

I’m watching now. Just need to remember every time you mention limit resets that I can ignore it. Well, I mean, I guess I get a limit reset on August 1. It’s just not fun when you use your allocation on the 3rd on the month (which I’ve done).

Same this happening even on Medium effort

if it does reduce token usage by 3-4x then its very useful.

$200k in tokens is a number I can't even process as a beginner lol, wild seeing how deep people go into optimizing usage on these subs.

Text walls are efficient for searching, but I get it. A video would definitely help visualize those prompt engineering tricks. Just hope the narrator doesn't ramble.

what are you even spending that many tokens?

Codex limit configs are the one case where I actually want the article. You end up ctrl+F-ing the same section three times mid-workflow.

when the model is actually cracked, usage caps become the biggest pain in the workflow

Then there is the @claudeai fable 5 users. We are just waiting 4 days to use our subscription again. Not hours.

The real unlock is knowing when to stop prompting and start editing files. Most people burn tokens asking for work they can steer in 10 seconds.

They nerfed the model, but at least we have weekly usage and it doesn't consume as much while being solid.

paying hundreds just to hit a brick wall every time is total insanity

Check this out — the final destination 😂

I am taking a day leave to spend all @thsottiaux

And Codex removing the 5-hour limit makes lot of sense now 🙂

these caps kill the dev flow faster than buggy code ever could

lee101/codex is the best way :) and have aliases you use to launch it at a set model/reasoning level and shift up down to change reasoning so always pick the best for the job. also helps you stack max plans or go auto thinking/automodel

gpt-5.6 unlocked and people are out here writing articles like its a tax filing 😂

Most useful part is the stop point. I’d make it a literal line in the task brief: plan only → pause; implement → PR → first review pass → stop. 5.6 going longer is great until “helpful” quietly turns into 15% of the window.

30-minute video for “this article could have been a video” is a very founder-coded bit.

the video was also a good one for those who haven’t given the article a read haha

Youre hitting limits before you need to because your usage patterns are still 99% sampling, not continuous queries.

Lmao this guy must be living life fast to burn that kind of tokens already!.

feels like it is time to switch to local instances to escape these arbitrary cloud quotas

Nice haircut

so youre just paying for the privilege of hitting rate limits faster. genius.

Good tips, but there's a bigger point hiding here: none of these harnesses (Codex, Claude Code, etc.) have real budget controls. "Stop points" and "lower your reasoning level" are just users doing manual rate-limiting because the tool won't do it for them. What we actually need is a token budget per task, not per reasoning level. Tell the agent "you have 500k tokens for this task, spend them however you want" — let it decide when to go deep and when to stay shallow. Right now we're tuning proxies (effort level, fast mode) instead of the actual constraint (cost). This is true across every agentic coding tool right now, not just Codex.

bro said article couldve been a video as a joke but also meant it fr codex tutorial content is speed running 2014 youtube era 💀

Burning $200k in tokens and still needing a guide to dodge the limit says the sub pricing was never built for how power users actually work. Metered billing would probably serve him better than a fixed monthly cap.