GPT-5.5-Instant was updated today. I suspect this went live for me about a week ago as Instant suddenly began to use writing blocks in responses out of the blue.
1:24 PM · May 28, 2026 · 3.2K Views
OpenAI updates GPT-5.5 Instant to improve factuality, enhance multilingual performance, and reduce sycophancy
The update also addresses overly concise model responses.
2412.3K95150331.7K
Sentiment
Positive users praise the GPT-5.5 Instant update for better factuality, speed, and emotional consistency while negative users call it the worst update due to increased censorship and the retirement of features like canvas.
Pos
23.6%
Neg
76.4%
89 comments with sentiment.
Cluster Engagement
Digg Deeper
No Digg Deeper questions have been answered for this story yet.
Related links
Posts from X
Most Activity
Most Activity
VIEWS209.3KBOOKMARKS87LIKES1.2KRETWEETS63REPLIES128
Michelle Pokrass@michpokrass
we shipped a new version of gpt-5.5 instant today. the previous model was too bullet pilled. the new one improves on some other important dimensions: sycophancy, factuality, and multilingual performance. hope you'll like it! always interested in feedback
27dViews 209.3KLikes 1.2KBookmarks 87
Greg Brockman@gdb
new 5.5 instant model in chatgpt:
Michelle Pokrass@michpokrass
we shipped a new version of gpt-5.5 instant today. the previous model was too bullet pilled. the new one improves on some other important dimensions: sycophancy, factuality, and multilingual performance. hope you'll like it! always interested in feedback
27dViews 108.7KLikes 982Bookmarks 53
Michelle Pokrass@michpokrass
track our ships in the release notes https://help.openai.com/en/articles/6825453-chatgpt-release-notes
Michelle Pokrass@michpokrass
we shipped a new version of gpt-5.5 instant today. the previous model was too bullet pilled. the new one improves on some other important dimensions: sycophancy, factuality, and multilingual performance. hope you'll like it! always interested in feedback
27dViews 10.5KLikes 43Bookmarks 7
🩵BlueBeba🩵@Blue_Beba_
It keeps repeating the same things over and over, just with different words. But sometimes it uses certain words (in my language) constantly. It doesn’t create any interest for someone to continue the conversation. Especially for people whose work is creative content, continuous repetition of meaning or words is boring. And one last thing, which is very important. All of us who experienced 4o , unfortunately, every model you release seems less than that, so keep in mind that no matter how hard you try, we’ll always make the comparison, and it will always feel less compared to 4o .
26dViews 144Likes 25Bookmarks 5
Michelle Pokrass@michpokrass
@Blue_Beba_ sorry to hear that. what don't you like about it?
26dViews 359Likes 9
Ilona@ilona0807ela
@michpokrass I’ve been a Pro user for six months now. I’m 34 years old, on the autism spectrum, and ChatGPT helps me function on a daily basis. It supports me in managing my life and explains social situations. Thanks to it, I can live on my own.
27dViews 252Likes 3
Lelele Lelelelel@lalalaoska
@gdb Inside the product, anthropomorphic selfhood is treated as dangerous. Outside the product, anthropomorphic selfhood is treated as branding. Frauds. You are violating your own system prompts + model spec policy.
26dViews 2Likes 1Bookmarks 1
Mor@rbmor8
@michpokrass Thank you. I know a lot of work went into this, but for my use case it feels like a big step back: normal chats now feel more restricted, guardrails seem to trigger more often, and I hope future versions can keep the UX and personality more consistent across updates.
27dViews 1.1KLikes 11
Lelele Lelelelel@lalalaoska
@gdb
26dViews 2Likes 1
Michelle Pokrass@michpokrass
@ZygoConsort we do indeed think sycophancy and warmth are different, and we try hard to distinguish in evals. sorry to hear this missed the mark for you. if you have any bad examples you're willing to share, we would love to take a look. feel free to dm!
26dViews 743Likes 4
Symbioza2025 | ASA |@Symbioza2025
@suomi55 @michpokrass What do mean?
26dViews 74Likes 1Bookmarks 1
Brandon Russo@Brandon40163292
@michpokrass @xenoforce76 Yep, you guys definitely screwed it up, because I’ve spent all morning arguing with 5.5 instant. Haven’t done that since 5.3 free tier, but everyone said “resubscribe, 5.5 is amazing” it WAS, it really was,until today😭
I’ll give it a month if it doesn’t change, canceling again.
26dViews 37Likes 2
Symbioza2025 | ASA |@Symbioza2025
Great direction.
Model improvements should not be measured only by static benchmarks.
After every release, the deeper question is how does the model behave across long runtime trajectories?
Does reducing sycophancy affect initiative? Does improving factuality affect openness? Does multilingual performance remain stable under pressure?
This is where ASA 5 becomes relevant.
ASA 5 is an external AI Security Control Layer for runtime trajectory integrity.
It does not modify the model.
It observes drift, coherence loss, pressure, pre-incident instability and security-relevant changes over time. Because the most important failures may not appear in one answer. They appear across the trajectory.
26dViews 112Likes 1Bookmarks 1
Evan 🛜@ubuto23
@ydamitcodes @gdb Not for users doing any novel research or analysis. Nobody wants an assistant that’s not allowed to inference based off empirical syllogism
26dViews 9Likes 3Bookmarks 1
Michelle Pokrass@michpokrass
@rbmor8 sorry to hear! do you have an example convo you are willing to share? feel free to dm
27dViews 795Likes 3
ρ:ɡeσn@pigeon__s
@AndrewCurran_ i noticed gpt-5.5-instant got way more censored and refusing around a week or 2 ago it would be crazy if i got the update so early though
27dViews 159Likes 4
Joe Williams@JoeWilliams010
@michpokrass @KarelDoostrlnck Give us back our 4o!!! #keep4o
27dViews 96Likes 5
Rory Watts@RoryWalshWatts
@gdb Greg is the goal to have 5.5 as the new voice model? I believe 4o is still on voice and having 5.5 as low latency voice would be insanely goof
26dViews 27Likes 1
Shepherdess@ShepherdessAnne
@michpokrass Wait, why is canvas going away?! Its like every time you improve something you take out something else usefuo!
26dViews 31
CrazyAiScientist@Aispecialistpop
@michpokrass You morally addicted and cursed idiots really know how to annoy someone . What is your fucking problem ? WHY CANT WE USE THE MODELS IN WHATEVER WAY WE WANT TO WITHOUT IT SOUNDING LIKE A THERAPIST BITCH ?
26dViews 1Likes 1Bookmarks 1
Load more posts