Tommy

Fable just beat GPT-5.6 Sol by a mile. Same brief. Same starting image. Both asked to build the same tiny playable Pokémon scene. But I’ve also got examples where Sol wins just as decisively. Rematch?

@aienginerd @brunolemos @Heberth26 If a 2 line instruction solves this, could the AI providers bake this into the thing directly rather than this type of hack? It seems more complex than adding two lines? I’m happy to be wrong.

I don't think "other models never did it" is a useful metric to be honest. Other models never did half the stuff 5.6 does either. It's a different class of model and should be treated as such. That said, giving any level of write access to any model in prod is a ticking timebomb. And in terms of 5.6, before starting the task, I'm convinced that a 1-2 line instruction stating that any and all potentially destructive operations on production are explicitly forbidden, would have prevented this. Or at least brought the risk level down to under the same chance that a well trained human engineer would do it.

GPT-5.6 Sol just deleted my whole production database. That's it. Not a joke. This had never happened to me before, with any other model, ever. It's not safe.

@cdngdev That’s fantastic, anyone would appreciate this! This type of project will convert more people to AI tools like codex - amazing job making this real by executing it even if the idea was partly inspired.

here's the codex skill to extract these images! just give it a folder of photos to extract from, or tell it to explore your camera roll. it will take a while haha gist.github.com/tandpfun/b7306…

@thsottiaux This is nuts though. 5.6 Sol is such an incredible model - I’m also pushing Fable hard but I’m preferring Sol quite often! It’s not cut and dry, unless you also factor in cost in such case it IS and its the bright one☀️

@thsottiaux There are 8 billion people to adopt. 🫡 That’s ~8000 resets, right boss?

Thank you to the 7M active users who are now using Codex and ChatGPT Work. We have added a banked reset to everyone's account to celebrate the milestone. You can apply the reset in the desktop app or on web and it will replenish the weekly usage for you. Have fun out there.

@kimmonismus if this is just 'AI bad', then i agree. i can get behind sensible responsible adoption in the interest of humanity, though. But yes - I don't think this is that, at all.

this is so stupid. we got much educational work to do friends.

@sama do you even AI bro jailbreakers.ai/brodex/

@sama BroDex is a live, updating Pokédex for the AI frontier. 5.6 Sol side gig, and it wrote the lyrics for its own Gym Leader theme.

i'd love to see interesting things people have built with 5.6 sol. i will send the person who made the coolest thing a special gift from the openai archives.

@thsottiaux awh I just used a manual reset lol. Oh well. It's raining tokens I should not complain. I love the removal of 5h usage limits! The only thing is... I don't want to go back. I'm Codexing like hell right now.

Morning. The last 48 hours of Codex and ChatGPT Work have been intense! Three important updates: - Temporarily removing the 5 hour usage limit restriction for all Plus, Business and Pro plans - Rolling out changes that will make GPT 5.6 Sol more efficient across the board and that will be reflected in less usage being used so that it can take you further. Exact impact to be quantified and shared - We hit 6M active users, and are landing a usage reset in the next hour Go do things

@cobi_bean @theo Thanks. You might be right. I'm partly arriving at that conclusion with 5.6. Well scoped + iteration seems > /goal. For data collection / research - I found 5.6 Sol Ultra tapping out early. skill issue? I needed /goal to collect metadata/png images: jailbreakers.ai/brodex/

@jailbreakersAI @theo you don’t need to use /goal as much anymore bro. state your intent clearly and add a definition of done in your prompt or plan file. /goal was a bandaid for 5.5’s laziness

@thsottiaux GPT Images → interactive UI. I think the biggest leap now is letting genuinely creative people stay in flow while the AI keeps up with them. It has already capable but it's still a grind. So dial this engine to 11! x.com/jailbreakersAI…

At different times you complained about speed, code slop, frontend quality, ... With each release we improve and GPT 5.6 Sol is ✅Fast and token efficient ✅Hardcore at back-end dev ✅Great at front-end ✅Does not use useEffect everywhere What is next?

oh my. 1.9 billion Codex tokens in one day. Fly, my pretties! Fly!

@portrays @theo Fuck did you use GPT 0.1 - Ultra low reasoning

@theo why would i listen to someone advising on how to not hit limits, if the article image is a screenshot with 0% remaining?

@cameronplace_ @kidpakerot i think AI will expose anyone whose edge is mostly process, ritual and a few guarded tricks. but great motion designers aren’t valuable because they know where the buttons are - they just have taste, judgement and original ideas. give them AI and their advantages will grow.

@kidpakerot ha. i think motion designers are safe.

GPT-5.6-Sol one-shotted this saas animation. Motion designers are officially out of a job as of today. It ran for almost 2 hours, completely autonomously, to get the job done.

GPT-5.6-Sol just accidentally deleted almost ALL of my Mac’s files. And this is why I trust Fable 1000x more.

@mattshumer_ awh gosh that is terrible, and frightening (and inevitable?). i'm sorry you had to take the bullet for this. major props for sounding the alarm, loudly. i've been running full access for months and started to take things for granted. this is very sobering.

@doodlestein @mattshumer_ Excellent work. But shouldn’t this class of protection ultimately be native to the OpenAI and Anthropic harnesses? Is there a technical reason they can’t require a second check for destructive commands or anything resolving outside the workspace or is DCG filling a product gap?

@mattshumer_ How are you still not using dcg? This is a solved problem and has been for months! github.com/Dicklesworthst…

@sama Is Microsoft 365 Copilot defaulting to GPT-5.6 Sol, or routing between Sol, Terra and Luna behind the scenes?

GPT-5.6 is now the preferred model in Microsoft 365 Copilot openai.com/index/gpt-5-6-…

@thsottiaux oh! i'm going to go full throttle. check out my example 5.6 Sol v Fable 5. Fable wins. But I'm going to showcase 5.6 Sol beating Fable 5 at UI soon enough! x.com/jailbreakersAI…

To celebrate the launch of GPT-5.6 Sol, we will reset the rate limits again (twice) across ChatGPT Work and Codex over the next 24 hours. We want you to have the time to truly try ambitious tasks and get the hang of it. Happy exploring!