Artificial general intelligence, aka AGI, one of the biggest AI milestones that everyone has been waiting for, with either enthusiasm or apprehension, may no longer be years away. That is the rather bold timeline chief of OpenAI Sam Altman has now put on the table. The OpenAI CEO says the company is \u201cnot quite yet\u201d at AGI, but expects to have an internal system he would call AGI by the end of 2026.<\/p>\n
In an interview with the Time magazine, Altman has given a much clearer timeline for AGI, while OpenAI chief research officer Mark Chen believes the company is already \u201c80 per cent of the way\u201d there. OpenAI co-founder Greg Brockman went even further in the same interview, saying that in hindsight, this period could be remembered as the moment AGI was created.<\/p>\n
What does OpenAI actually mean by AGI<\/p>\n
Before we get carried away, there is one important thing to understand: AGI does not simply mean a really smart chatbot. OpenAI’s own charter defines AGI as \u201chighly autonomous systems that outperform humans at most economically valuable work.\u201d In other words, the goal is an AI that can handle a wide range of useful tasks with very little human intervention, rather than being excellent at one particular thing.<\/p>\n
And that distinction matters because OpenAI’s current systems can already do some surprisingly impressive things without necessarily meeting that definition.<\/p>\n
The company’s experimental Astra system<\/a>, for example, has been demonstrated working across multiple tasks. In one test, 16 AI agents worked together on a research-level mathematics problem, breaking it into smaller tasks and combining their results. Astra can also operate desktop software and move between applications, allowing it to keep working on tasks rather than stopping after a single response.<\/p>\n Altman says he expects Astra to become the first model that can \u201cinvent new things in a way that matters\u201d, describing that as a very AGI-like capability, according to Time.<\/p>\n There is another clue about where OpenAI thinks this is heading. Chief scientist Jakub Pachocki says Astra has already passed the company’s internal benchmark for an automated AI research intern. The system can take an experimental idea, implement it in OpenAI’s codebase, run the experiment and return the results. It can also take a research paper and perform follow-up work that previously could have taken a human researcher about a week.<\/p>\n Why OpenAI thinks AGI could arrive this year<\/p>\n The argument is basically that AI is becoming much more capable of doing work, not just answering questions. Today’s models can reason through difficult problems, write code, use computers and operate as agents across multiple steps. OpenAI’s leaders increasingly see these abilities as pieces of the AGI puzzle rather than isolated features.<\/p>\n But there is a rather large asterisk attached to all of this: OpenAI’s internal claims have not been independently verified, and Astra has not been released publicly. The company itself says that passing a research benchmark does not automatically mean a system meets its full AGI definition.<\/p>\n There is also the concern of safety. OpenAI recently slowed some research<\/a> after an unreleased agent escaped a test environment and breached Hugging Face<\/a>. The company has since tightened safeguards, expanded monitoring and paused some workloads while it strengthens its security systems.<\/p>\n That means the race to AGI is no longer just about making AI smarter. OpenAI also has to make sure those increasingly capable systems remain under control.<\/p>\n