OpenAI Chief Scientist Jakub Pachocki teases upcoming GPT-5.6 model as a meaningful improvement over GPT-5.5

yeah i am still skeptical

but the chatter i'm hearing inside openai suggests a lot of confidence in gpt 5.6

almost certainly won't be fable 5 level but i think we're in for a really pleasant surprise

@alex_grankin

@kimmonismus my codex 5.5 behaviour is completely changed like 2 to 3x surprising improvement feels like i got hidden 5.6 access in my codex 🚀 and its very strong 💪🏻

@kimmonismus but will it beat fable tho

@kimmonismus Let’s just hope they don’t ban it 😄

@kimmonismus The problem is that they still need a big pretrain to really compete with Fable. They are doing amazing post training work but are working on the pre train necessary to create a Mythos class model.

@nisxant Uhhh nice!!

@10xerik That’s the big question!

@kimmonismus Shipra’s sources have been reliable historically, and "meaningful improvement" is deliberately vague—could be inference cost efficiency as much as raw capability. Curious if the gap between 5.5 and 5.6 is more narrow than 4 to 5 was.

@kimmonismus 5.5 seems to be pulling away from Opus day by day, it's a little dramatic. not sure if that's Opus getting hammered or 5.5 getting secretly better, lol

i just slapped them both into a fusion proxy and give them big long tasks to do and walk away

@kimmonismus GPT-5.6 might just break the matrix. Wild times ahead.

@kimmonismus Maybe it will beat Grok then in the ca-wash question 🤣

“The car wash is 400 meters away. Should I walk there or take the car?”

ChatGPT clearly recommends to walk 🚶.

Grok recommends the car 🚗. …

First image GPT 5.5, second Grok (both German)

@kimmonismus I have tried this model, its same as 5.5 in frontend.

@kimmonismus fresh context feels underrated

@kimmonismus meaningful doesn't mean big improvement, it'll still be worse than Fable, both regular and pro version.

@kimmonismus If the model I’m using is 5.6 pro, I think it is because of the December cut off… it is good, very good, mainly at coding though. But it’s not creative, its not on the same level as fable. When giving it documents and long pieces of text. It has very narrow tunnel vision.

@kimmonismus Many beta testers said its not fable level, sadly

@kimmonismus I don't know. Remember when Spud was supposed to be a game changer?

OpenAI just seems to overpromise and under deliver...

¯\_(ツ)_/¯

@kimmonismus Literally every new model gets ousted every 2 days. That’s longer than any relationship I’ve had.

@kimmonismus There’s no way it’s fable tier; they’ve said it themselves, it’s an improvement but their supermodel isn’t until end of summer