Altman expects AGI by year's end — on OpenAI's terms
Chief research officer Mark Chen puts OpenAI at 80 percent of the way there, while the Astra models already work as an automated research intern.
A research lab at night, desks empty and monitors still glowing — illustration, not a press photograph.
Sam Altman expects OpenAI to have an internal system by the end of 2026 that meets the company's own definition of AGI — a claim no one outside OpenAI has been able to test.
At a glance
- Altman expects an internal system by the end of 2026 that he would call AGI — internal, not a public release.
- Chief Research Officer Mark Chen puts OpenAI at 80 percent of the way to AGI, an in-house estimate with no outside test.
- Chief Scientist Jakub Pachocki says Astra handles unattended what takes a human researcher roughly one week.
- OpenAI's AGI definition: highly autonomous systems that outperform humans at most economically valuable work.
- Puck-shaped hardware is slated for early 2027; ChatGPT ads target the 92 percent of users on the free tier.
Sam Altman expects OpenAI to have an internal system by the end of 2026 that he would classify as artificial general intelligence. Internal is the operative word: nothing about a public release was promised. The remarks come from a TIME report published on August 26, 2026. Altman also concedes the company is not quite there yet.
What the leadership team is claiming
Chief Research Officer Mark Chen puts the company at 80 percent of the way to AGI. Co-founder Greg Brockman expects this stretch of time to be read later as the period when AGI arrived. Both are in-house assessments, not results from an outside evaluation.
Astra as an automated research intern
The Astra model family carries the claim. Chief Scientist Jakub Pachocki says it already writes code inside OpenAI's own codebase, runs experiments on its own and reports back with results. Work that would occupy a human researcher for roughly one week is handled unattended. Altman expects Astra to be the first model that, in his words, actually invents new things in a way that matters.
The same capability underpins what OpenAI calls persistent agents — colleagues in software form that carry a task across days rather than answering one prompt at a time.
The definition is doing the work
OpenAI defines AGI as highly autonomous systems that outperform humans at most economically valuable work. Whether Astra clears that bar is settled by the company's own reading, not by an independent test. Apply a stricter definition — full substitution for human knowledge work, say — and the same facts support a different verdict. The headline rests on the yardstick as much as on the model.
What else is on the roadmap
A small puck-shaped device is slated for early 2027, sensing its surroundings and driven by voice. Alongside it: a humanoid robot and ChatGPT Work, which folds the Codex coding agent into ChatGPT. Advertising inside ChatGPT is also planned, aimed at the 92 percent of consumer users who never pay. Under the label sponsored agents, brands would present their own AI experiences.
What is still unverified
None of this can be checked from the outside yet: Astra is unreleased, no third-party benchmarks exist, and the 80 percent figure comes from OpenAI itself. The two sources read for this article are the German and English editions of the same newsroom and trace back to the same TIME report; we have no second, independent confirmation. Until a model can be tested outside OpenAI, the AGI promise remains an announcement.
FAQ
What did Sam Altman actually say about AGI?
That OpenAI should have an internal system by the end of 2026 that he would call AGI under the company's own definition. He said nothing about releasing it.
What is the Astra model family?
A model system that, according to Chief Scientist Jakub Pachocki, writes code in OpenAI's codebase, runs its own experiments and absorbs tasks worth about one week of human research work.
How does OpenAI define AGI?
As highly autonomous systems that outperform humans at most economically valuable work. That wording comes from OpenAI, not from an independent body.
