OpenAI previews sandbox agent workspace for GPT-6 Astra
The preview gives Astra its own isolated environment with persistent files and resumable execution, turning it from a question-answering model into an agent that can work over long periods.
holy sh*t. openai just gave GPT a real workspace. the new sandbox agents preview turns GPT-6 Astra from a model that answers prompts into an agent that can actually keep files, run shell commands and continue work across sessions. instead of rebuilding context every time, each agent gets its own isolated environment with scoped permissions, persistent state and resumable execution. [this is the part that makes long-running agents actually usable] 1/ give the agent a workspace 2/ limit exactly what it can touch 3/ let it resume after the process stops that’s a much bigger shift than another model upgrade. GPT is starting to look less like a chatbot and more like infrastructure.
This post quotes another post: x.com/gippp69/status/2097696163424014406