— GPT-5 is the first moment where you see a glimmer of AI doing new science. It’s very tiny things, but here and there someone’s posting like, “Oh, it figured this thing out,” or “Oh, it came up with this new idea,” or “Oh, it was a useful collaborator on this paper,” Altman said.
GPT-5 is very capable
GPT-5 was deemed a very capable research assistant in the Olympiad on Astronomy and Astrophysics, but not good enough yet for autonomous research.
It has also won gold in the Math Olympiad and topped the International Collegiate Programming Contest, and upon those events, OpenAI researchers said the days of benchmarking is over and that the next frontier is scientific discovery.
«GPT-6 can really do it»
There is ample anecdotal evidence of GPT-5 helping solve tricky mathematical problems, and many researchers have found it quite useful, especially in saving time on complex issues — but GPT-6 will take it to the next level, Altman says:
— There is a chance that GPT-6 will be a GPT-3 to 4-like leap that happened for Turing test-like stuff for science, where 5 has these tiny glimmers and 6 can really do it.
It should also be the first AI that could be running an entire company, eventually:
— Well, but eventually it should get to the whole thing, right? So we can use this and then try to work backwards from that. I find this a very interesting thought experiment of what would have to happen for an AI CEO to be able to do a much better job of running OpenAI than me, which clearly will happen someday, Altman opined.
He was then asked when we can expect GPT-6 and said «not this year.»
Massive investment, massive compute
OpenAI has been investing some $1.4 trillion for over 40 gigawatts compute power in preparation for future ChatGPT models, with much of it expected to come online in the second half of 2026.
Solving scientific puzzles will almost certainly be in contention with this level of computing power and expected gains in capacity.
ChatGPT-6 should also be a better functional chatbot, Altman said earlier, with better memory and more personalization.
The only timeline we know of so far is Altman saying it should take less time than going from GPT-4 to 5, which would put it at roughly two years out. GPT-5 was launched in August.
Read more Conversations with Tyler, transcript. Video is here.