Third Moment Research reports running 10,000 AI agents in extended research sessions
The AMW Read
The reported concurrency and session duration meaningfully extend the agent deployment baseline, but a single customer account without cost or reliability evidence supports limited downstream conclusions.
Third Moment Research reports running 10,000 AI agents in extended research sessions
Third Moment Research uses OpenAI models and Codex to run as many as 10,000 agents concurrently on quantitative research, with sessions lasting 36–48 hours. The company says Codex helped reduce a backtest from about three hours to three to five minutes. Its system also generates synthetic scenarios to improve agent performance. These are company-reported results presented in an OpenAI customer story; the supplied account does not include independent validation.
The market significance lies in coordinating large numbers of agents over extended research sessions. This places the deployment within the agent market's central execution question: whether autonomous workflows can produce useful results reliably beyond short demonstrations. The reported backtest improvement gives a concrete example of faster execution within that workflow. It does not establish the quality of the research, the reliability of the full agent system, or the economics of running thousands of agents concurrently. Synthetic scenarios add another evaluation question: whether improvements carry over to the research tasks the system ultimately performs.
For builders and investors, the concrete implication is to evaluate task-level gains separately from system-level performance. A comparable deployment would need to show that shorter backtests preserve result quality, alongside completion rates, total operating costs, and human oversight across extended sessions. Agent count and session duration describe deployment scale; those additional measures would establish whether that scale delivers repeatable value.