we've been testing some new methods for improving writing quality. you may have seen @sama's demo in late march; GPT-5-thinking uses similar ideas it doesn't make a lot of sense to talk about better writing or worse writing and not really worth the debate. i think the model writing is interesting, novel, highly controllable relative to what i've seen before, and is a pretty neat tool for people to do some interactive fiction, to use as a beta reader, and for collaborating on all kinds of projects. the effect is most dramatic if you open a new 5-thinking chat and try any sort of writing request for quite some time i've wanted to let people feel the agi magic I felt playing with GPT-3 the weekend i got access in 2020, when i let that raw, chaotic base model auto-complete various movie scripts and oddball stories my friends and I had written for ~48 hours straight. it felt like it was reading my mind, understood way too much about me, mirrored our humor alarmingly well. it was uncomfortable, and it was art base model creativity is quite unwieldy to control and ultimately only tiny percents of even ai enthusiasts will ever try it (same w the backrooms jailbreaking that some of you love). the dream since the instruct days has been having a finetuned model that retains the top-end of creative capabilities while still easily steerable all reasoning models to date seem to tell when they're being asked a hard math or code question and will think for quite some time, and otherwise spit out an answer immediately, which is annoying and reflects the fact that they're not taking the qualitative requests seriously enough. i think this is our first model that really shows promise at not doing that and may think for quite some time on a writing request it is overcooked in certain ways (post training is quite difficult) but i think you'll still like it 😇