OpenAI has hinted it could reach artificial general intelligence (AGI) within this year. It has not declared it has achieved AGI, but internally it judges it could soon secure a system close to AGI.

On Aug. 26, local time, blockchain outlet Cryptopolitan reported that OpenAI CEO Sam Altman (샘 알트먼) said in an interview with Time that OpenAI has not “fully reached” AGI, but “will soon have a version close to it.”

Inside OpenAI, some assess that substantial progress toward AGI has been made. Chief Research Officer Mark Chen (마크 첸) said current progress is about 80 percent toward AGI. Co-founder Greg Brockman (그렉 브록먼) said the current period could later be judged as the time when AGI first appeared.

OpenAI’s confidence is underpinned by a new model line called Astra. OpenAI introduced Astra last year as a system that mimics an automated research assistant. Chief scientist Jakub Pachocki (야쿱 파초키) claimed Astra has already reached the level it originally targeted.

Astra aims to do more than simply answer questions. According to Pachocki, if a user presents an experimental task, Astra can write the required code in OpenAI’s internal software environment, run the experiment and retrieve the results.

There was also an explanation that if a research paper is provided as input, it can carry out the research task within a week. This shows the potential to develop into a “persistent agent” that can perform tasks over long periods without humans directing every step.

The capability OpenAI ultimately wants to verify through Astra is not simple automation. The core test is whether it can actually create new knowledge, beyond organizing existing information or finding already known answers.

Altman said at a customer preview event that Astra would be “the first model to invent something new in a way that really matters.” He added, “That is a very AGI-like thing.”

This is interpreted as meaning OpenAI views not only autonomous task execution but also the ability to make new discoveries and inventions as important in judging AGI. It envisions a virtuous cycle in which AI conducts research on its own and uses the results to develop better AI.

Some also view autonomous research AI as still far from AGI. Key issues are whether AI based on large language models (LLMs) can actually generate new discoveries, and whether it can generalise knowledge across tasks in diverse fields rather than being limited to specific research.

Another variable is that the definition of AGI itself is unclear. OpenAI defines AGI as “a highly autonomous system that outperforms humans at most economically valuable work.” Researchers differ in the standards they use to judge AGI.

Ultimately, OpenAI’s 2026 AGI target is expected to be assessed alongside debate not only over real advances in AI technology but also over how far the definition of AGI should extend. Whether Astra can advance to the level of independently creating new knowledge is expected to be a key point that will determine the possibility of an OpenAI declaration of AGI.