UhuBuhu

@Jobox05 @flash_canadian There is No big News openai ist stopping sora to get more compute to Not fall behind on the llm Side since Google and Claude catched Up the only real News is that other labs are making faster Progress on AI than openai and people think this is a sign of the ai downfall

@flash_canadian can you point me to the news?

Watching the downfall of AI is so satisfying

@nicdunz didn't they announce an update was coming or wasn't there leaks about that voice mode 1.5 or something like that

chatgpt voice is so good now. what did they do?? why havent they said anything about it??

@alexgrama @kimmonismus depends in my opinion it shows that llms still lack a rly basic part about reasoning but how relevant that part rly is i don't know we will see that in the coming years

@kimmonismus is this really a relevant test? it's absolutely fantastic at what matters. who the fuck cares about this stupid ass question? same with the strawberry test and all the other retarded ones before that.

ChatGPT 5.4 still doesnt get it.

@thsottiaux I still don't get why you Name the Models Codex the cli Codex and the web Thing Codex Name the Model gpt 5.3 Code or coding

Our naming team has been cooking

@emollick For me Codex was a bigger Leap than o3 cause sonnet was already pretty Close at least in coding

From an AI user perspective, the four big leaps so far in ability: 1. GPT-3.5 (ChatGPT, November 2022) 2. GPT-4 (Spring 2023) 3. Reasoners (starts with o1-preview, but the real deal was o3, Spring 2025) 4. Workable agentic systems (Harness + good reasoner models, December 2025)

@JJBalls9 @jpschroeder Smarter but slower at least on the 20€ plan i use both

@jpschroeder Codex 5.3? I so I may give it a try.

Sonnet 4.6 and Opus 4.6: Anthropic trained Sonnet 5, and it almost outperformed Opus 4.5. With some RL on benchmarks, it could even outperform 4.5, but it was now dramatically smaller (cheaper). So? They renamed Sonnet 5 → Opus 4.6. But what of Sonnet? They distilled "Sonnet 5" into an even smaller model and rebadged it Sonnet 4.6. So now both models are just a fraction of their original size and cost to run. Even better, it left a little bit of extra compute overhead, which can be sold for 6x the cost as "fast mode." The models are not *actually* better than what they replace but...margin. I can't blame them too much for that.

@OP__Nico @Angaisb_ I am fine already got chatgpt fpr codex and claude for opus. The free tier pf gemini gives me everything i need from gemini but ty for the tip

@kecksbe @Angaisb_ As a subscription is really valid tho! And u get really generous 4.6 opus use in antigravity with your subscription! Give it a try, to me is really the ultimate deal :)

Time to go back to Plus Pro was amazing while it lasted

@OP__Nico @Angaisb_ The problem with gemini is that claude is worse in anzigravity and that codex is smarter than claude but if dev isn't your main focus its a great deal

@Angaisb_ Man since I switched to gemini I don't think I can go back to anything you get really good limits within the UI + Antigravity use. Antigravity opens up doors to opus 4.6 too, which I had running for at least 4 hours yesterday building Swift code. It's the ultimate deal really

@adonis_singh Why call it spark and not mini though

achieves about gpt-5.3-codex-low levels of accuracy (at xhigh itself) at a noticeably faster way faster and smarter than the previous 5.1-codex-mini model though, which is a bigg win

finally!! up to 1000 tps speeds on a smaller version 5.3 codex!

@thsottiaux The terminal ui sucks in claude code i get a way better overview of what the model is actually doing in codex especially back when the model was slow i often questioned myself is the model working or got it stuck? But overall the model is great maybe try to speed it up even more

What could we do better on Codex? App, model, strategy and features… what’s wrong in how we approach things that we should improve immediately?

@adonis_singh i want to use 5.2 instant for quick easy searches but the model beeing this much worse than the thinking and even 4o defeats the whole purpose of that if i need to write 5 prompts for it to do what i want

@adonis_singh i feel like its always dumber simple example i asked both for the current state of a Tournament named jbb 5.2 instant told me 3 times that i probably mean jbbl. I told it no search for jbb searched again for jbbl after 5 attemts i went to 4o it instandly seachred for jbb by itse

gpt-5.2 instant is dumber than gpt-4 at times

@erawrlyne @ThePrimeagen Disagree it eats way more tokens for the same task

@ThePrimeagen Feels that way with 4.6 tbh

it would be funny if 5.3 is just 5.2 and it's an experiment to see how much psychosis there really is

@testingcatalog @patience_cave

BREAKING 🚨: CLAUDE OPUS 4.6 IS ROLLING OUT ON THE WEB, APPS AND DESKTOP! TESTING TIME 🔥

@KryptoWolfGER Sowas von live

Gehen wir live oder geben wir auf? #Bitcoin

@patience_cave @Angaisb_ Maybe all hope in you isn't gone CHAIR

@Angaisb_ big

No Sonnet 5 today Patience Cave won this time

@kimmonismus I am missing the new gemini agentic flash which is in my opinion the best one right now and mixtral ocr which was the best one before gemini flash agentic vision. Paddle ocr is good metric but not on mixtrak or gemini lvl

So we got SOTA OCR with just 0.9B params. GLM-OCR is a lightweight (0.9B params) multimodal OCR system built on the GLM-V encoder–decoder stack multimodal OCR system built on the GLM-V encoder–decoder stack. Love it!

@TheRealSynetos @camsoft2000 We need claude code ui with a router that actually works routing between opus and codex. Oh and codey needs a speed boost

@camsoft2000 We just need the cozy feeling of Claude Code UI/UX mixed with Codex capabilities and that's it. I just love how CC feels to use. And also we need Codex models that are better at frontend. Or if anyone has a workflow to make better frontend, I'm begging you to share it please

Having used Claude Code with Opus 4.5 a lot recently, I can tell you that for me, Codex CLI with GPT-5.2-Codex wins. It's a relief to go back to Codex; it feels like home. Not saying CC is rubbish, it's just Codex gets it done, Claue Code is too lazy, replies to my queries without re-checking code, and requires more steering and planning. Codex just doesn't need all that. I'm sure some will disagree, and it's subjective for sure, but I much prefer Codex to Claude Code. It suits the way I work.

@apoorvdarshan @theCTO Slow but smarter

@theCTO isn't it slow?

i miss the real Opus 4.5 rip legend

@GodswillRed @sharmasarkastic Bro are you a girl?

@sharmasarkastic Left is a true man and right is a boy.

Girls left or right?