OpenAI expects AGI internally this year

A TIME feature published August 26, based on more than two weeks of interviews with over 20 OpenAI executives, employees, investors, and rivals, reports that Sam Altman is convinced OpenAI will have an internal system he would call AGI by the end of 2026. Chief Scientist Jakub Pachocki told TIME that Astra, an unreleased model, has already met the company’s internal benchmark for an “automated AI research intern”: taking an experimental idea, implementing it in OpenAI’s own codebase, running the experiment, and reporting results, and using a research paper to carry out follow-up work that previously took a human researcher about a week. Chief Research Officer Mark Chen put OpenAI’s overall progress toward AGI at roughly 80 percent.

Every part of this is worth reading with the source clearly in view: it is OpenAI’s own internal assessment of an internal, unreleased system, not an independently verified result, and it does not mean a public AGI product is coming this year. Altman himself reportedly told TIME the company was “not quite yet” there.

Read the claim, not the headline

The concrete piece of this worth tracking is narrower than “AGI”: whether Astra’s specific research-intern capability, propose, implement, run, report, holds up once independent researchers can test it, similar to how Astra’s earlier claim of solving ten open math and CS problems got real scrutiny once it shipped, and how OpenAI later paused Astra over a cyber-capability threshold rather than release it unchecked. The AGI framing is a marketing and morale story following a stretch of high-profile OpenAI departures; the research-automation capability, if it holds up, is the part that would actually change how research teams work.