Illustration: LittleDot concept art, AI-generated.
GPT-6 is still in use, while messages about GPT-7 have already started circulating. I clicked through, of course. After reading a few reports, though, I wanted to know something more basic: who first said this, and what exactly did they say?
One concrete source I could trace was a September 25 set of conversation notes published by Fireside Alpha. It introduces Boris Power as OpenAI's head of applied research and records him saying that roughly 80 to 90 percent of OpenAI's research is aimed at GPT-7, GPT-8, and models beyond them. AI.info later reported on the same conversation.
That statement is worth paying attention to. But what I found was a description of research direction, not a GPT-7 release date or a product configuration sheet. The percentage is an approximate estimate made in conversation; it does not mean a project is "90 percent complete."
I also checked OpenAI's API model catalog. As of my review on October 9, GPT-7 was not listed there. At least for now, I cannot turn this research news into "GPT-7 is available to use."
What I care about is how a new model reaches people
I am not surprised that a lab is already working several generations ahead. The harder question is whether research first becomes a new model, or first shows up in the tools we open every day. Changes to chat, coding, and API access do not necessarily reach users at the same time.
For someone like me, who uses AI to build websites, revise articles, and investigate problems, the model name is only the entry point. The practical questions come next: can I continue the project I already started? Will confirmed requirements stay in context? When I switch tools, do I have to explain everything all over again?
In an earlier piece about Tibo's Day 3, I wrote about speed and continuity between projects. Seeing GPT-7 mentioned makes me hope even more that the products we already use will become steadier. A new release can be exciting, but it cannot finish today's half-done work for us.
Of course I am willing to wait for a better model. If it takes fewer wrong turns and understands complicated requirements more accurately, that is worth looking forward to. But before people can actually use it, I do not want to celebrate early or schedule projects around a rumored date.
I will keep following the news while carrying on with the work in front of me. When the model really appears in a tool, I can try it on a complete project: read the materials, make changes, and check the result. Then we will not have to guess from a headline how far GPT-7 has come.