OpenAI Finally Reveals GPT-4.5 & 5 Official Roadmap. What to Expect
If have been using OpenAI’s ChatGPT, you may likely have found the long list of the different models rather annoying or at least annoying. Here are some examples of the models in the picker dropdown;
- GPT-4o (great for questions)
- GPT -4o. with scheduled tasks
- o1 -uses advanced reasoning
- o3- Mini; Fast at advanced reasoning
- o3- mini-high; great at coding and logic
- o1-promo de – best at reasoning
- GPT-4o mini; fast for most questions
- GPT-4 legacy model.
Deciding which one is best for what kind of work can be confusing, especially if are new to the space. It is often the case that a company will too many product offerings will likely fail because consumers can easily be overwhelmed by this.
It seems Sam Altman -the CEO of OpenAI- is finally realizing this fact and wants to do better. In his recent statement on x, on the release of the official roadmap for GPT 4.5 and GPT-5, he says”
“We want AI to “just work” for you; we realize how complicated our model and product offerings have gotten.”
When someone wants to use ChatGPT, they generally don’t know which one to choose from.
GPT-5 as an All-in-One System
Instead of having several standalone models as shown above, OpenAI is exploring the less complicated route of having an All-in-one model with the GPT-5.
“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 idea is that this model will be using a soft of unified machine intelligence. The design will likely be such that uses a RouteLLM structure or something similar.
Basically, the AI takes users’ prompts and sends them through an analyzer to evaluate the question/prompt based on certain criteria like cost, latency, QA, etc. It determines what type of LLM (or models) will best suit the question, and then uses the LLM Router to send it to them for processing. And you get a response.
“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.”
The user doesn’t have to worry about what model to use. Simply enter the prompt and you leave the rest for the machine to deal with. Of course, it will eventually depend on whether you have some access to some functionality (paid subscription services).
The last Non-Chain-of-Thought Model as GPT-4.5
Sam Altman also mentioned that the last non-chain-of-thought model (the old method of training LLM) will be GPT-4.5, which internally is called Orion.
And this is their next model, before the release of the GPT-5 (the all-in-one model).
When is GPT-5 going to be released?
When asked when the release of these models will be, Sam Altman indicated “weeks/months” as his response.
It seems reasonable enough to speculate that GPT-4.5 will be in the first quarter of 2025, and GPT-5 by the start of the second half of the year.
There is also no reason to doubt that these releases will happen a lot faster, as things often do in the AI space.
Free-tier users will have access to GPT 5
Free-tier users will still have access to GPT-5 when they release it. This will be at the standard intelligence setting.
“The free tier of ChatGPT will get unlimited chat access to GPT-5 at the standard intelligence setting (!!), subject to abuse thresholds.”
You need to pay a subscription (Plus subscribers) to access GPT-5 at a higher level of intelligence.
Furthermore, the Pro subscribers (about $200 at the moment) will be able to run GPT-5 at an even higher level of intelligence. These models will incorporate voice, canvas, search, deep research, and more.
See : OpenAI in Play: Musk’s Surprise Bid for His Former Creation
