OpenAI Counts on Paid Users to Test ChatGPT-4.5
Today in a livestream, OpenAI finally announced the launch of its GPT-4.5 model, but with one twist: it currently requires a $200 per month ChatGPT Pro subscription to use it.
This is because the new Large Language Model, or LLM, is technically still in a state of “preliminary research.” Apparently it’s short for “please pay us for the privilege of being a beta tester.”
Jokes aside, the company promises that GPT-4.5 provides more “natural dialogue” and performs better when it comes to programming, pattern recognition, writing and “practical problem solving.” Please note that GPT-4.5 is not intended to be a reasoning model, meaning that it will not have the self-correction or deep exploration capabilities of OpenAI o1 , o3-mini , or the future full version of o3. In the leaked document, the company said it doesn’t consider GPT-4.5 to be a “cutting-edge” model, but it is OpenAI’s largest LLM, which should make it ideal for day-to-day queries as the large database of training materials should facilitate faster responses.
According to OpenAI’s internal tests, the new model hallucinated about 24.7% less often than GPT-4.0 and was about 34.3% more accurate. About 57% of internal human testers were reported to prefer GPT-4.5 over 4.0.
In other words, GPT-4.5 isn’t exactly on the cutting edge of artificial intelligence, but it should be an improvement over other non-reasoning models, providing a nice middle ground between cheaper, more traditional LLMs and their often paid successors. The only major omissions appear to be multimodal features such as AI voice mode and video input, although OpenAI says it is working on a user interface update, which seems to hint that some version of these capabilities may come to the model in the future.
This makes sense given that it’s technically an early release and OpenAI says it’s “still learning” the limitations of GPT-4.5 and how people can use it. To this end, LLM is planned to roll out to all paid ChatGPT plans over the next two weeks before presumably reaching free users as a replacement for GPT-4.0 once it comes out of preview.