OpenAI targets AGI by 2026 using its new Astra model

OpenAI leadership believes the company is on the cusp of achieving artificial general intelligence (AGI), with CEO Sam Altman predicting an internal system will meet his definition of the milestone by the end of 2026. Chief Research Officer Mark Chen estimates the firm is already 80 percent of the way there. At the center of this push is Astra, an upcoming family of models designed to function as an automated research assistant. According to Chief Scientist Jakub Pachocki, Astra can already take an experimental concept, write the necessary code, run the trial, and report back, compressing a week of human research labor into a fraction of the time.

Astra is designed to support persistent agents that can operate independently over long periods. Altman expects this system to be the first capable of inventing meaningful new things, potentially triggering recursive self-improvement. Alongside these software advances, OpenAI is developing physical hardware. The company is working on a pocket device, a wearable, and a puck-shaped tabletop device that uses voice mode to interact with its environment, slated for release early next year. Altman also confirmed plans to build humanoid robots.

To fund these massive computing demands, OpenAI is shifting toward tighter product monetization. A project internally dubbed "The Merge" combines the Codex coding agent with ChatGPT to create ChatGPT Work, an always-on agentic platform designed for the mass market. Additionally, the company is introducing advertisements to ChatGPT to monetize the 92 percent of consumer users who do not pay for subscriptions. These ad efforts include "Sponsored Agents," which direct users into interactive, brand-hosted AI environments.

For developers and enterprise practitioners, these shifts signal a transition from passive chat interfaces to highly autonomous, persistent workflows. If Astra and ChatGPT Work deliver on their promises, technical teams will transition from writing code and running manual tests to managing fleets of digital coworkers that execute complex, multi-day research and development tasks independently.