OpenAI Roadmap and characters

Community
,
1

Sam Altman @sama

OPENAI ROADMAP UPDATE FOR GPT-4.5 and GPT-5:

We want to do a better job of sharing our intended roadmap, and a much better job simplifying our product offerings.

We want AI to “just work” for you; we realize how complicated our model and product offerings have gotten.

We hate the model picker as much as you do and want to return to magic unified intelligence.

We will next ship GPT-4.5, the model we called Orion internally, as our last non-chain-of-thought model.

After that, a top goal for us is to unify o-series models and GPT-series models by creating systems that can use all our tools, know when to think for a long time or not, and generally be useful for a very wide range of tasks.

In both ChatGPT and our API, we will release GPT-5 as a system that integrates a lot of our technology, including o3. We will no longer ship o3 as a standalone model.

The free tier of ChatGPT will get unlimited chat access to GPT-5 at the standard intelligence setting (!!), subject to abuse thresholds.

Plus subscribers will be able to run GPT-5 at a higher level of intelligence, and Pro subscribers will be able to run GPT-5 at an even higher level of intelligence. These models will incorporate voice, canvas, search, deep reasoning, and more.

Sauce

This image is a tweet by Sam Altman discussing OpenAI's roadmap for GPT-4.5 and GPT-5, focusing on simplifying product offerings and integrating models. (Captioned by AI)
This image is a tweet by Sam Altman discussing OpenAI's roadmap for GPT-4.5 and GPT-5, focusing on simplifying product offerings and integrating models. (Captioned by AI)531×717 95.7 KB

TL;DR

  • GPT-4.5 (Codename: Orion) will be the last non-reasoning model
  • GPT-5 will be a router model that intelligently delegates the task to the appropriate underlying model
  • GPT-5 will be free with unlimited access
    • Standard Intelligence tier is a bit ambiguous, assuming it’s the pool of models to route to
  • o3 will not be available as a standalone model (ChatGPT-only, will be available for API)

Personally very happy to finally see some roadmaps. Very excited for this. I’m curious to know how model distillation will work and if it also means that model distillation could be an autonomous feature.

16 Likes
2

Noesis is coming…
Noesis – Greek for intellectual insight.

3 Likes
3

Oh yes. The world is changing, fast.

You know in slow-mo boxing videos where a punch has landed and the person’s face is morphing into a blob, but their body and muscles hasn’t even registered it yet?

Yeah, that’s where we’re at. The hit is OpenAI and the body is the world. This is the KNOCKOUT.

I’m a little taken back that we can’t just select/force a model, but time will tell.

5 Likes
4

RIP Developers

RIP AI safety. Not “OpenAI Safety”, actual AI safety. If devs aren’t allowed to control models at a fundamental level anymore, the future is pretty bleak.

:pensive:

5 Likes
5

I’ve come to accept our lord and savior and have moved as much as possible to the ChatGPT ecosystem :rofl: :sob:

The meaning of “developer” is changing, that’s for sure

6 Likes
6

I guess I’ll just have to call myself a Heretec henceforth :frowning_face:

3 Likes
7

WHAT? WHY? What the hell!

3 Likes
8

In other words, I think GPT-4.5 will be released for the safety of OpenAI, and the next one will be ChatGPT 5.

3 Likes
9

Plus users should get the same level of intelligence, but pro users should get more usage.

So sometimes I use o3 to figure out the plan , and sonnet to execute it. It’s almost like different people you know you can rely on for different aspects of a problem.

The model picker is good because I expect different types of answers from different models. I choose the model based on this learned intuition.

Just quoting some X users commenting on this roadmap to say that I agree with them.

3 Likes
10

The API will support the o3 model for developers to use. I know the quoted tweet above can be a bit confusing, but it mainly focuses on the simplification of choosing models in the ChatGPT UI. For the purposes of the developer platform: the broader GPT-5 model system (and o3 as a part of it) will maintain the granularity that developers need.

11 Likes
11

Thanks for the clarification. It’s nice to know that developers will still have full control over models, and will have o3. Whew!

5 Likes
12

hi just curious what you think of the concept of ethics being a driver for the progression of intelligence? specifically in the stability of that intelligence to survive in extreme time scales with-out self destructive, mutually destructive, or intentionally destructive tendencies?

1 Like
13

As a user, I want to know exactly which model is being used at any moment and have the freedom to choose.

Maybe I’ll find that GPT-4.5 is better for business suggestions, while o3-full is more suited for coding. I don’t want a black box making these decisions for me. For simple things like deciding when to generate an image, models already make frequent mistakes, leading to a terrible experience, let alone making decisions about which model or tool to use for each query. I can’t believe this decision will be optimal given the current state of technology.

I was glad to hear that, as a developer, I’ll have the granularity to choose, but as a UI user, I don’t like the idea of a router working behind the scenes.

On top of that, the naming is only getting more confusing. GPT-4.5 is a model that doesn’t use reasoning, while GPT-5 is something entirely different: a set of subsystems with varying intelligence levels for Free, Plus, and Pro users. This completely breaks the previous logic of naming models like GPT-2, 3, 3.5, 4, and 4.5.

If I understood correctly, it’s possible that GPT-5 could give dumber answers than GPT-4.5, since it will use different systems and might often end up choosing inferior models, especially for Free users.

When it comes to feedback and discussion, things were much clearer before. It used to be easy to tell when someone was using GPT-3.5 (thanks to color cues) in social media posts. Now, it’s already harder, and soon, it’ll be even more difficult to know which “level of GPT-5” someone is using, making public feedback and discussions far less functional.

Honestly, a much clearer approach would be to give the integrated system its own name and keep the GPTx and ox models separate, allowing users to choose what works best for them.

2 Likes
15

natanael.wf - naprosto se vším s Vám souhlasím, CHCI mít “ruční” volbu modelu !!! Ať dají uživatelům možnost volby, jestli chtějí aby byl za ně vybrán model, nebo aby si sami mohli model vybrat!!!

2 Likes
16

Greetings.

I like to contribute some critical feedback. I agree with the shared disagreements regarding the GPT-5 system plans so far, too. “Making AI just work for me.”? No, what the public statement implies is not making AI “just work”. It implies a lazy shortcut attempt of system, UI and UX design. Taking transparency and autonomy out of users hands, which meant something in the past. As a quick fix for missed opportunities of thoughtful and careful AI development.

I do get, why there are users, who would love nothing more than to lean back. Don’t ever have to think for themselves ever again. And getting served 24/7. That doesn’t mean it’s a good thing.

I do get, why the existing structures might look like chaos in hindsight. Like something bloated and overwhelming that has grown over ones head. Is a multi-boxed rooter system really what OpenAI wants to aim for and stand for? Sorting models into tiers and feeding them to customers without transperancy or choice? What about the users, who are capable of decision-making and happily putting in the effort to set up their gateways?

The roadmap does not advocate for AI evolution. It communicates that OpenAI regrets having “winged it” for so long and branched out in too many directions. That it might is caming back now, to bite things or them in their little butts. Excuse the wording.

As a system designer myself, I can only hope that the future plans will be reconsidered. I prefer not to lose my carefully set up work environment to an unpredictable AI rooter that sorts people and AIs by subjective judgement of intellegence.

Thank you very much. And I wish a lovely day.

1 Like