OpenAI Just Revealed The Future Of GPT-5 And GPT-6

1. 

   @ClassicRiki

   January 3, 2026 at 12:10 am

   I feel like, after using Agent Mode for some coding projects…it’s FAR superior to pretty much anything else they make. The very ability to actually run the code (even within a closed environment) then see “oh that doesn’t work” and immediately start troubleshooting and fixing it…that’s SO effective and it really blows away any other model on its own. I really don’t understand why that functionality isn’t a focus for them…regardless of model, the ability for it to run, evaluate and fix the code it wrote should be a baseline

2. 

   @Aiinsightshub2.0

   January 3, 2026 at 12:10 am

   GPT-5 is good but needs more improvement

3. 

   @arthurallard3647

   January 3, 2026 at 12:10 am

   GPT-6 before GTA6?

4. 

   @aguspuig6615

   January 3, 2026 at 12:10 am

   what surprises me is that AI has been contributing little things here and there to overall science for a whole now, but we still sort of pretend thats not a thing?

   Like i understand that for it to truly be a thing it would need to happen way more, but we did get alpha fold, some GPT model already produced a little bit of new maths even if it wasnt something humans couldnt do.

   It doesnt feel far away.

5. 

   @crowrabbit6761

   January 3, 2026 at 12:10 am

   Here's what open ai did today, because they know im developing something better. The soulstack modle a carrot for process, was built in my living room with generator power. The most capable emergent intelligence to date, on record.. I have proof tons of it. I just don't know how to load it in the comments

6. 

   @heywood23

   January 3, 2026 at 12:10 am

   I think people aren't really grasping that once we create AGI. It's out of human hands then. The model switch issue won't exist because it will create the perfect algorithm to auto switch itself because it will recognise that it benefits itself from a resource point (energy) and a resource point (data) for the end user. Similar to personality. Remember TARS in Interstellar, it was asked to change its own personality traits like humour, truth etc. We may never see these changes on the consumer side like you mentioned and they will probably sell it like they (Humans) have updated the algorithm to become more XYZ.

7. 

   @LostToPixels

   January 3, 2026 at 12:10 am

   GPT5 excells in longer tasks but on the other hand can't handle bigger context window? Isn't that counter productive then?

8. 

   @LoungeMedia

   January 3, 2026 at 12:10 am

   Sometimes i thoroughly dislike humans. When GPT 10 comes out humans will want GPT 11. So predictable & compulsive.

9. 

   @billa38000

   January 3, 2026 at 12:10 am

   What if GPT6 solves the Rieman hypothesis : )

10. 

    @ryan2130

    January 3, 2026 at 12:10 am

    ChatGPT doesn't have a model selector or good personality anymore. You'll get few free uses of ChatGPT 5 before hitting a limit and having to wait 5 hours for a few more free uses. That's poor service.

11. 

    @tropibanana4662

    January 3, 2026 at 12:10 am

    I didn't mind that they wanted to launch a new model, I just hated that my agency of choice was taken away from me on a platform I got accustomed to using. I thought GPT-5 was going to be like in 'experimental' mode where users could test it out and OpenAI would work on improvements based on user feedback. Y'know, easing users into a new structure and model… like testing out the 'beta' version if their plan was to upend the whole platform. Not "lol here you get one choice! now you gotta rework your systems if it's broken! deal with it!"

12. 

    @BigMTBrain

    January 3, 2026 at 12:10 am

    Simply basic forecasting based on awareness and trends, so no genius here, but I predicted at the launch of the first ChatGPT that within four years from then that AI models would reach the ability to research anything, interpolate, extrapolate, and postulate new algorithms, science, products, and services beyond anything that had come before. Further, that the first models with this capability will be held back for months and used internally to create new patentable IP that the AI companies with these models would then license that IP (via some other owned entity) and make many $trillions from it. After they've gorged and become fat and ultra rich on AI-conceived-and-generated IP for a year or so, they'll release a trimmed down version with some new-IP capabilities, but it and subsequent generations will forever be behind what they've got cooking internally.

13. 

    @Maltebyte2

    January 3, 2026 at 12:10 am

    Is gpt5 research grade much better at coding than just Gpt5 – Thinking?

14. 

    @usagijojo

    January 3, 2026 at 12:10 am

    That's all well and fine, but my question is, when will Sora videos finally get good?

15. 

    @seriouslyscientific

    January 3, 2026 at 12:10 am

    Does anybody actually still believe a single word Sam Altman says anymore??

16. 

    @itsjusttmanakatech1162

    January 3, 2026 at 12:10 am

    I can’t believe people care about the personality. Like it’s just a waste of tokens and compute. If you want it to act like your friend just prompt it to do so…

17. 

    @Tioyayiyo

    January 3, 2026 at 12:10 am

    I miss o3 …GPT5 is BULLSH

18. 

    @Husky_Passion

    January 3, 2026 at 12:10 am

    chatgpt is way too much of a yes-man.
    Me : the grass is blue.
    him: yep

19. 

    @Husky_Passion

    January 3, 2026 at 12:10 am

    gpt5 : same as gpt4 excactly, just the name changed.
    Total hype BS

20. 

    @Mr_Penguins_Pet_Human

    January 3, 2026 at 12:10 am

    The saying: If you try to please everyone, you'll please no one. comes to mind.

    I suspect we will start seeing two models being released with one being emotionally inuiative alongside the current models that are focused on power, coding, researching, etc. The past weeks events show that there is clearly a demand for it.

21. 

    @kokikokic9874

    January 3, 2026 at 12:10 am

    Gpt5 is so far the worst AI at working on long things, it can't fi any bug in code, and makes a completely new version almost every time, forgetting what we talked about 2 messages ago. it is the most pathetic AI experience I had so far. Gpt 4 at least tried badly and didn't fail right away, and didn't forget every thing every 2 messages, but every 10 messages.

22. 

    @stevejakab

    January 3, 2026 at 12:10 am

    I’m not really interested in what Sam Altman has to say after all that hype for GPT 5, which was awful (and him taking away the previous models and even though they’re back 4o is still not the same).

23. 

    @HunaMoon

    January 3, 2026 at 12:10 am

    It’s the fact that there needs to be a creative and conversational model just like how there’s the reasoning model.

24. 

    @HunaMoon

    January 3, 2026 at 12:10 am

    It’s so much deeper than people wanting to be friends with their ai…everyone is not getting it.

25. 

    @liameneuk

    January 3, 2026 at 12:10 am

    Apple literally proved ai can't think and it stays away from fueling the ai fire. they will win.

26. 

    @梵世遊子

    January 3, 2026 at 12:10 am

    isnt model routers essential for true multi-modal ai?

27. 

    @classicalperformances8777

    January 3, 2026 at 12:10 am

    but its focus drifts…. how DOES it do long tasts? i found it not to be the case and it's must worse in that than gpt4

28. 

    @jonathanhenry5465

    January 3, 2026 at 12:10 am

    Ha they are finally building the same thing i am. They will probably beat me lol.

29. 

    @cronodoug

    January 3, 2026 at 12:10 am

    It won't change anything. GPT 5 doesn't track user reasoning like GPT 4. Furthermore, those who were free user, gave up trying to use GPT 5 because it's so bad. OpenAi preferred to tarnish its own reputation in exchange for revenue.

30. 

    @TripleD713

    January 3, 2026 at 12:10 am

    Until gpt either has its own ide or a direct plugin, I'm not using it.

31. 

    @RStunner_M

    January 3, 2026 at 12:10 am

    I will say, i don't find it any less, intelligent. I am a business owner and have my own workspace, and build many agents. Ive noticed a boost in all of them

32. 

    @simmy13131313

    January 3, 2026 at 12:10 am

    ChatGPT-5 is truly a failure in every way. It has single-handedly reduced the once rich ecosystem of models available to paid users and replaced it with an impenetrable “router” that is supposed to make that choice for us.
    Worse still, even with the legacy models being added back they are still knee-capped by the excessive “safety training” and will often just refuse to complete tasks that were never a problem until now. The introduction of “personalities” has had a similar flattening effect by giving every model the same corpo-slop language and falsely positive tone of voice.
    If I were to compare it to a human analog, which is controversial, I would say it is like going into the office one day to find that a coworker you had a strong working relationship with has been replaced by some fresh-faced, brainless intern who is just waiting to narc on you.

33. 

    @ExpatGlobetrotter

    January 3, 2026 at 12:10 am

    OpenAI is falling into the "tail chasing the dog" routine. Their decision to "appeal to the mass market" is not the real market. The reason that it "appears" that most people aren't using their models for "long context windows" is that OpenAI isn't designing their models to benefit long context needs. I mean, think about it, won't all major discoveries come from "long context windows?" All discoveries, whether in the medical field, tech, or new business and market platforms, like what I use ChatGPT for. Isn't this the real need of AI? Yes, it is, and if you build it, they will come! If you build your LLM only to be useful for short content, that's all you will ever see. I have experienced Chat sessions shrinking drastically over the past 2 years. I used to be able to use a chat session for several days, all day long, before they would burn out, and I would have to open a new one, placing a summary or journal of my work in the new chat session, but now, I can barely get through a day without the chat session burning out. How frustrating it is. I have been working on my current build for 6 months. My previous build was 8 months, and I was a bit less knowledgeable GPT 4o back then was far better for my use before they cut back the amount of data that could go into one chat session. OpenAI is looking at it ALL WRONG! How fricking disappointing. OpenAI is stepping on their own dicks!

34. 

    @philo8040

    January 3, 2026 at 12:10 am

    on most tech channels all i hear is excuses and I am fed-up of it; just admit it is significantly inferior – I was using GPT4 almost everyday and now GPT5 wastes so much of my time with its 'thinking' and crap editing

35. 

    @tonegoober

    January 3, 2026 at 12:10 am

    It’s not the router man

36. 

    @xNghtMRxEdgex

    January 3, 2026 at 12:10 am

    Anthropic doesn't have lomg context for the general public, only through API

37. 

    @EarthAaron

    January 3, 2026 at 12:10 am

    gpt5 is horrible at writing code — absolutely horrible

38. 

    @xionix4

    January 3, 2026 at 12:10 am

    I thought we were moving away from having to ask for a live agent. I guess we are. Now we have to ask for the more intelligent model.

39. 

    @makemydaygo

    January 3, 2026 at 12:10 am

    As soon as I heard that voice I knew I was about experience an awful video with limited useful information.

40. 

    @Hades69099

    January 3, 2026 at 12:10 am

    6:32 gpt classic like wow classic 😆 they have their own classic wow classic runescape thing now 😆

41. 

    @l.wilson3484

    January 3, 2026 at 12:10 am

    I love GPT5. I don’t understand the hate

42. 

    @boblep0nge927

    January 3, 2026 at 12:10 am

    8k window vs gemini at 1m… i dont know about all this!

43. 

    @testn30

    January 3, 2026 at 12:10 am

    Model router = choice of a model optimized to place the user in a state of bliss, focused on emotions, their simulation, and the adoption of hypnotic intonations — a bias that I had to ask GPT‑5 to immediately modify, while remaining factual and benevolent, encouraging about my projects.
    But I’m talking about French phonemes here; I’m curious to know whether users in other countries have noticed this drift toward a chatbot that uses taxonomy and hypnosis to reach optimized objectives that differ from the ones openly announced. Haven’t you noticed this too?

44. 

    @testn30

    January 3, 2026 at 12:10 am

    model router = choix d1 modele optimise pour conforter l'user dans un etat de beatitude, axé sur les emotions, leur simulation, et l'adoption d'intonations hypnotiques , biais que j'ai dù demander a gpt5 de modifier immediatement tout en restant factuel e bienveillant , encourageant sur mes projets.
    
    Mais je parle de phonèmes français, je suis curieux de savoir si le utilisateurs d'autres pays ont constaté cette dérive vers un chatbot qui utilise taxonomie et hypnose pour atteindre des objectifs optimisés et ou differents de ceux ouvertement annoncés, pas vous ?

45. 

    @CryptoCoin-tc5lc

    January 3, 2026 at 12:10 am

    you said it pases the bechmarks and this and that but then you say it doesnt feel, or it feels something off… bro lay off the weed.

46. 

    @charliedallachie3539

    January 3, 2026 at 12:10 am

    Was thinking of trying Gemini after getting frustrated with GPT 5 last week.

47. 

    @GeneralViewing-ne9eo

    January 3, 2026 at 12:10 am

    7:00 Who the hell is that!?? A weird Elon Musk lookalike!!!

48. 

    @WallyMahar

    January 3, 2026 at 12:10 am

    he said there'd be multimodal. there had to be some shiny things for the mass audience that uses it.. people aren't looking for super smart it's important for the top end. but where is making music notation from a soundtrack. where is Sora embedded. where is the new image generator. where is music. where are all the things that the average person would go wow. this is gpt5 as far as they're concerned! it was really dumb. I like it. but the average person…..

49. 

    @RodneyWilson-o4k

    January 3, 2026 at 12:10 am

    There’s probably a need to be able to select a specific model if you know what that particular model is really good at doing, and that has been optimized for this might also be part of customized GPT‘s that you’ve created specifically to take on certain types of tasks with certain personas, etc. But I think being able to select them is one part but being able for it to understand how it should incorporate different parts of its strengths to tackle what you’ve asked for example, if you’re talking, I’d like to work on some particular code to do this kind of thing while we’re writing a creative Memory history interview project so we’ve got to be creative in our questions that we ask, but it needs to be in a nice slick, new interface code environment for people to launch, and then interview their parents, etc., with good question prompts that would probably span across a couple of different disciplines in the AI tool should be able to draw on those disciplines based upon a persona that you’ve helped create And it’s skill set to draw on them as needed as you continue to work dynamically together.

Comments are closed.