Need smarter insights in your inbox? Join our weekly newsletters to get solely what issues to enterprise AI, knowledge, and safety leaders. Subscribe Now
Within the 2.5 years since OpenAI debuted ChatGPT, the variety of giant language fashions (LLMs) that the corporate has made accessible as choices to energy its hit chatbot has steadily grown.
The truth is, there at the moment are a complete of seven (!!!) totally different AI fashions that paying ChatGPT subscribers (of the $20 Plus tier and costlier tiers) can select between when interacting with the trusty chatbot — every with its personal strengths and weaknesses.
These embody:
- GPT-4o
- o3
- o4-mini
- o4-mini-high
- GPT-4.5 (Analysis Preview)
- GPT-4.1
- GPT-4.1-mini
However how ought to a person determine which one to make use of for his or her specific immediate, query, or process? In any case, you’ll be able to solely decide one after the other.
The AI Impression Collection Returns to San Francisco – August 5
The following section of AI is right here – are you prepared? Be a part of leaders from Block, GSK, and SAP for an unique have a look at how autonomous brokers are reshaping enterprise workflows – from real-time decision-making to end-to-end automation.
Safe your spot now – house is restricted: https://bit.ly/3GuuPLF
Is assistance on the best way?
Assist seems to be on the best way imminently from OpenAI — as experiences emerged over the previous few days on X from AI influencers, together with OpenAI’s personal researcher “Roon (@tszzl on X)” (imagined to be technical crew member Tarun Gogineni) — of a brand new “router” perform that can routinely choose the most effective OpenAI mannequin to answer the person’s enter on the fly, relying on the precise enter’s content material.
As Roon posted on the social community X yesterday, July 20, 2025, in since-deleted response to influencer Lisan al Gaib’s assertion that they “don’t need a mannequin router I would like to have the ability to choose the fashions I exploit”:
“You’ll nonetheless have the ability to choose. It is a product to guarantee that docs aren’t caught on 4o-mini”
Equally, Yuchen Jin, Co-founder & CTO of AI inference cloud supplier Hyperbolic Labs, wrote in an X submit on July 19.
“Heard GPT-5 is imminent, from somewhat chook.
- It’s not one mannequin, however a number of fashions. It has a router that switches between reasoning, non-reasoning, and tool-using fashions.
- That’s why Sam mentioned they’d “repair mannequin naming”: prompts will simply auto-route to the proper mannequin.
- GPT-6 is in coaching.
I simply hope they’re not delaying it for extra security exams. 🙂“
Whereas a presumably way more superior GPT-5 mannequin would (and can) be enormous information if and when launched, the router could make life a lot simpler and extra clever for the common ChatGPT subscriber.
It will additionally comply with on the heels of different third-party merchandise such because the web-based Token Monster chatbot, which routinely choose and mix responses from a number of third-party LLMs to answer person queries.
Requested in regards to the router concept and feedback from “Roon,” an OpenAI spokesperson declined to supply a response or additional info at the moment.
Fixing the overabundance of selection downside
To be clear, each time OpenAI has launched a brand new LLM to the general public, it has diligently shared in both a weblog submit or launch notes or each what it thinks that specific mannequin is nice for and designed to assist with.
For instance, OpenAI’s “o” collection reasoning fashions — o3, o4-mini, o4-mini excessive — have carried out higher on math, science, and coding exams because of benchmarking exams, whereas non-reasoning fashions like the brand new GPT-4.5 and 4.1 appear to do higher at artistic writing and communications duties.
Devoted AI influencers and energy customers could perceive very properly what all these totally different fashions are good and never so good at.
However common customers who don’t comply with the trade as carefully, nor have the time and funds accessible to check all of them out on the identical enter prompts and evaluate the outputs, will understandably battle to make sense of the bewildering array of choices.
That would imply they’re lacking out on smarter, extra clever, or extra succesful responses from ChatGPT for his or her process at hand. And within the case of fields like drugs, as Roon alluded to, the distinction might be considered one of life or dying.
It’s additionally attention-grabbing to invest on how an automated LLM router would possibly change public perceptions towards and adoption of AI extra broadly.
ChatGPT already counted 500 million energetic customers as of March. If extra of those folks have been routinely guided towards extra clever and succesful LLMs to deal with their AI queries, the affect of AI on their workloads and that of your entire international economic system would appear more likely to be felt way more acutely, making a constructive “snowball” impact.
That’s, as extra folks noticed extra good points from ChatGPT routinely selecting the best AI mannequin for his or her queries, and as extra enterprises reaped larger effectivity from this course of, an increasing number of people and organizations would seemingly be satisfied by the utility of AI and be extra prepared to pay for it, and as they did so, even extra AI-powered workflows would unfold out on this planet.
However proper now, that is presumably all being held again somewhat by the truth that the ChatGPT mannequin picker requires the person to A. know they actually have a selection of fashions and B. have some degree of knowledgeable consciousness of what these fashions are good for. It’s all nonetheless a manually pushed course of.
Like going to the grocery store in your city and observing aisles of cereal and totally different sauces, the common ChatGPT person is presently confronted with an overabundance of selection.
Hopefully any hypothetical OpenAI router seamlessly helps direct them to the proper mannequin product for his or her wants, after they want it — like a trusty shopkeeper displaying as much as free you out of your product paralysis.
