iyda

Most of this is nonsense, but Opus 4.8 just yeeted my entire DevDrive D drive and then filled my entire 2TB NVMe with who knows what. This happened with Auto mode on, and I honestly have zero clue what it did, since it was using ULTRACODE with dynamic workflows. Yes, I have backups because I don’t trust Claude and never will. This is precisely why GPT 5.5 is superior, my GPT 5.5 has never screwed me over this badly. Before anyone says, “oh, you shouldn’t let it run with so much power or full power,” sure, but again, it was on AUTO mode, so it had restrictions. I’m sure a “ultracode XHIGH” mode would think, hmm, deleting an entire drive is maybe not the solution to the task, hmm? Give me my GPT 5.5 back, please. Never again. Every time I’ve used Opus, it’s done some dumb crap with git reset or git checkout and lost hours of work (yes, I know I should commit often, use worktrees, blah blah, cool).@theo have ur laugh at this garbage... what a fucking joke

@codespicious Gotta add some more gradients and sparkles

@notiyda Yes 😅

Stackoverflow if it was vibecoded.

@dayvough Do I maybe use Codex a little too much?

Codex all day, everyday

@maria_rcks @theo @simpsoka Sorry maria <3

@theo @simpsoka get wrecked

@simpsoka I have to pick up on my usage, can't have you this close to my lifetime tokens

@theo @simpsoka Gotta set up ur game Theo come on get it to 100 billion

Google keeps releasing new models, and while they may be cheaper than most Frontier models, they do not perform as well and are not very good for coding. I am surprised the latest models were not better at coding, especially since they had anti-gravity and all the data from Opus 4.6. Still, they did not deliver. Gemini is still a mediocre model compared to anything from Frontier Labs, which is surprising for a roughly $4.6 trillion dollar company. The usage and rate limits on these models are also fucking ass. How can you tell me that Google’s own model, which uses all their optimisations and their own TPUs, has worse limits than Claude and OpenAI, which have to buy their compute? I’ve only used Gemini Models for front-end development, and even then it was about 50/50 whether it would do a good job or be an AI slop. It’s definitely strong at animations, and that’s one area where I’ve consistently given it praise. It’s usually been better at animations than most models, even if that may not be true anymore compared to Opus 4.8 and GPT-5.5 From what I’ve seen, Google’s priority and focus are not on today, tomorrow, or the next few months. They’re focused on building something that will become the standard in 10 to 20 years.

It really depends on what you’re trying to do. Claude is a better fit for more natural work and everyday workflows, especially if you don’t need to do a lot of heavy coding. If you’re doing a lot of coding, I’d recommend Cursor so you can use Composer 2.5, since it’s designed for higher usage than Claude.

Where should your $20 go? 💸 Cursor Premium vs Claude Premium

In this situation, it depends how much money you have. If you have a fuck tonne of money, go with Opes 4.8 because you can easily deploy thousands of subagents in a dynamic workflow. If you’re on the $200 plan, go with GPT 5.5 because it’s more consumer-friendly, cost-efficient, and token-efficient. You’ll get more usage out of it and probably do more.

If you had to recommend ONE AI Model to developers: GPT 5.5 or Opes 4.8

Honestly, OpenAI is moving in the best direction with token efficiency. The $200 plan lets more users use their models more, with higher usage limits, which is amazing. Even if their models are not saying they are right now (because they are very good), the price-to-performance difference is still worth using GPT over Anthropic models. Google models are a fucking joke, and the same goes for most Chinese models. They are all garbage compared to the frontier models, and I guess we’ll see what happens with SpaceX AI.

Cold take on what comes next: - OpenAI will flourish - Anthropic will continue to be profitable - Google will not catch up to Anthropic or OpenAI - no chinese company will catch up to Anthropic or OpenAI - the highest tier of intelligence will become a luxury product that only companies and multi-millionaires/billionaires can afford - most of the companies that invested massively in them will have massive returns - SpaceX’s AI will be fine and on par with Google by end of year - Nvidia will become the first 10T company

@T3chFalcon No, no, no, no. This is not true. It's definitely the clouds in the sky that hold my data.

Yup there's no cloud.

@Anaya_sharma876 Again, completely situational, and this is fucking pointless to say.

Be honest As a developer, Which one is better SQL or noSQL ??

@hiarun02 Who would've guessed that it would be Anthropic? Kuda made it 210 mil if you just use Codex.

Half a billion gone

A friendly reminder: you can disable the workflow keyword trigger in the Claude Code settings if you don’t want workflows to trigger when you say the word “workflow.”

Wait, did Anthropic implement a system where, if you hit your rate limit while using dynamic workflows, it automatically starts up again the second your rate limit is working? I just hit my 1-hour rate limit, and my workflows basically stopped. Nothing was happening. The second my 5-hour rate limit reset, it started working again instantly. Yo, that’s insane.

@rezoundous No, I think you're absolutely right.

@notiyda hahahaha. I must not be using that right

My personal ranking 1. Codex GPT-5.5 2. Claude Opus 4.8 3. Composer 2.5 4. Manual code 5. Gemini

@ashokgelal Honestly, it kind of reminds me of the situation between Intel and AMD.

lol. Hard to believe CC and Opus was the king not a long time ago.

@nicdunz You're kind of wonky.

did not know this was a thing, was just saying random word

@nalinrajput23 I dislike that the application comes with goofy-ass crypto shit.

" Name an app with 0 haters " i'll go first:

@karthikponna19 Which git command does Claude use the most? git commit. Which command do I hate that it uses the most? git reset and git checkout.

which git command do you use the most ?

@kimmonismus Is Windows finally gonna be good and not fucking shit? I can’t lie, they need to go ahead and get CodexGPT 5.5 and do a /goal on that bitch, then tell it to optimise Windows.

The 2026 collaboration i did not see coming.