OpenAI's write-up describes how Third Moment Research, an AI lab, operates 10,000 agents for more than 36 hours at a stretch. The agents are built using GPT-6 Astra and GPT-6.1 Sol, alongside Codex and Daybreak, the latter aimed at cybersecurity work.
According to the source, the lab's agents are designed to focus for days and improve with use. The framing suggests that long-horizon agentic behavior, rather than quick single-turn tasks, is the intended use case.
Since this is the only source, there is no independent or contrasting account to compare against. All details come directly from OpenAI's own description of the lab's work.