IuCC

@AhmedAlNeaimy @OpenAIDevs @sama @thsottiaux Yeah, I've been looking for answers too but haven't seen anything official yet. A ton of people seem to be running into this exact issue right now.

Guys @OpenAIDevs @sama @thsottiaux i can confirm Codex is eating the usage limits so fast, there is something wrong with all my 4 Pro 20x accounts! I even didn't use them heavily cuz i was in a travel, back to work yesterday and today and runs out the entire week limits.

@pawelk411 @jturntdev @thsottiaux @sama @OpenAIDevs Damn if thats the case, that sucks. No clue why no one that works at OpenAI is informing us about anything?

@IuCCtweets @jturntdev @thsottiaux @sama @OpenAIDevs yes, cache is bugged, limits are worse, compaction is bugged too. its an issue

OpenAI have secretly adjusted our limits. Last week before limit reset. I was using Xhigh all day. 5 day straight i couldn’t get my usage below 55% weekly usage. Since Yesterday, I’ve done 40% of my quota, out of nowhere. So whats going on ? @thsottiaux @sama @OpenAIDevs

@jturntdev @OpenAIDevs @OpenAIDevs @sama a lot of people are having this issue, is there something going on?

Still no response from The @OpenAIDevs As someone who literally uses Codex more than 99% of people. Something has to be off? I trust yous. But clarity would be great.

@Hub4Nerds @nexxeln @thsottiaux So strange indeed, I have the exact same problem

@nexxeln And crazy token eager lately @thsottiaux This morning 2 prompts, nothing unusual, 5.5 medium not fast..no tokens in 1 hour

why is gpt-5.5 being so dumb rn

@IamARoosterMAN Glad it's not just me then. Usage seems to be draining way faster than it used to.

@IuCCtweets Something seems extremely off. Seems like it’s draining at 10x the rate. I’m almost out on day one, where I’ve never been below 80% in a week I don’t think.

Is it just me or am I hitting my Codex usage limits way faster than usual recently?

@perkmaybe Hi!

Bro to Bro: build your x account Just say “Hello” and gain 700 mutuals here.

@GeorgeMayer Yeah, that’s probably the bigger issue honestly. A mediocre PR is annoying, but autogenerated "research" that still forces you to rebuild all the context yourself doesn’t really save much time either.

@IuCCtweets Totally, i will also say though on the tickets where it declined a PR, it left research and that was also unhelpful, basically had to do it all over.

Ran codex on a /goal to pull down every backlog ticket and open a PR if it was confident it understood the problem fully, otherwise leave questions/comments on the ticket. It worked for 11hrs. Some stats, observations and model usage: Over 75 tickets, it issued about 40 PRs. I'm halfway through the review and: 1) it probably 1-shotted 40%-ish so far. These are simple tickets. Mostly minor feature requests or cleanup 2) Probably had to pull down and update another 40% 3) Outright closed about 20%. Of that about 1/10th were fully misunderstanding the ticket and 1/10th were slop PRs that we just didn't need. Observations: 1) It's kind of predictable what it can one-shot. A classifier here would probably make this system work pretty well. 2) But it obviously cannot self-classify. it's hit-rate is below 50%. 3) When it misses, it's kind of a pain to the point of almost useless. Need to rebuild the context myself, pull it down, etc. 4) It actually did a great job with duplicate tickets. Properly resolved them to single changes, etc. This surprised me a bit because it burns through it's own context window quite fast. The tickets themselves also naturally cluster temporally, so that probably has something to do with it. 5) It did not do a good job with stale tickets/things that had already been fixed. Often would add more scaffolding / more random error handling. Just slop on top. Can probably improve guidance here. But just given a ticket it will assume it's open. 6) Misunderstanding ~10% of the tickets entirely makes this approach a bit untenable to me because it makes me more cautious/slow with the other ones. Model usage: Worked for 11hrs. Consumed ~40%-50% of my weekly codex (max). Tbh, they reset the limits so often that right now using /goal in these ways seems worth it. I also used fast w/ xhigh. So high token burn. Probably fine to use medium for this w/o fast and just work for 18 hours instead.

Whoever at @X designed this modal should probably reconsider their job choices

@cursor_ai So cool to see the scores on these benchmarks!

Introducing Composer 2.5, our most powerful model yet. It's more intelligent, better at sustained work on long-running tasks, and more reliable at following complex instructions. For the next week, we’re doubling the included usage of the model.

@jturntdev So strange, I see quite a lot of people complaining about it today

@IuCCtweets Me too

@wojciechGryzio Great job💪! Small start or not, getting your first real users and paid subscriber is a huge milestone.

Two weeks ago, my app got its first free trial. I didn’t post about it then. Today, the dashboard shows 2 active trials and 1 active paid subscription! It’s still a very small start, but it feels great to see real people using the app :) #buildinpublic #indiedev #solofounder

@joerossi98 @thsottiaux @OpenAIDevs I noticed this as well today

Am I the only one seeing more Codex usage today? It seems like it's burning through everything @thsottiaux @OpenAIDevs

@xdrewmiko @waelnassaf Honestly, great job! It's very intuitive

@IuCCtweets @waelnassaf Spent about a week on the website bro!

Drop your project URL ⬇️ You never know who might DM you!

@YotamBlu @iamkarank5 I kind of agree with you, but GPT-5.5 is also really strong for coding. Frontend/UI generations can still be hit or miss, but the backend code it writes is genuinely solid and honestly feels on par with the latest Claude models.

@iamkarank5 claude opus for coding, no question. gpt is solid for prose and reasoning but claude just gets the vibe of what you're trying to build. i've shipped way more with claude than i ever did spinning on gpt prompts.

As a developer which one is worth it for coding? - GPT-5.5 - Claude Opus 4.7

@YashHustle_22 Both Claude and Codex have their pro's and con's, but I switched to Codex due to Anthropic's decisions lately.

Vibe coders, What made you switch from Claude to Codex?

@viktoroddy Woah, this is insane!

Kimi K2.6 is actually a beast for design. ❤️‍🔥 Just recorded a tutorial on how to use Kimi K2.6 + to build award-winning $10k websites (step-by-step).

@mary_talkss @waelnassaf Thats cool!

@waelnassaf CursorClip.com, a tool for quick demo recordings with a one time payment instead of a subscription.

@xdrewmiko @waelnassaf Such a cool website! Love it!

@waelnassaf databerry.app