# Third Moment Research Uses GPT-6 Astra in 10,000-Agent System By Simon Yoon Canonical URL: https://www.tokenpost.com/news/technology/41286 Published: 2026-10-26T20:11:48.000Z Updated: 2026-10-26T20:11:48.000Z Section: Technology > The four-times speed comparison covers cited Financial Modeling World Cup challenges. Other computer-use scores and simulated task times measure separate results. Third Moment Research is using GPT-6 Astra in a multi-model system with capacity for 10,000 concurrent agents, putting the model into sustained research workflows in quantitative finance. OpenAI announced the deployment Oct. 14, naming GPT-6 Astra, GPT-6 Sol, GPT-6.1 Sol and Codex as components of the system. It can run 10,000 agents concurrently, with research sessions lasting 36 to 48 hours. Early applications include quantitative-finance research and tests of trading strategies against historical data. Its control layer can rerun incomplete work, end a task or escalate it to a human reviewer. The Financial Modeling World Cup comparison puts Astra at about four times the speed of the winning human competitor on the cited challenges. The result concerns those contest tasks alone; it does not establish speed across workplace software generally. Astra scored 72.6% on OSWorld 2.0’s v2026.08.08 offline set, compared with 65.7% for GPT-5.6 Sol. OSWorld 2.0 tests computer-use workflows and awards partial credit for incomplete task progress, so the percentage does not show what share of tasks the model finished end to end. Separate latency simulations put an Astra task at about 40 minutes, compared with about 75 minutes for Sol; those are modeled durations, not a workplace productivity study. Astra also scored 62.7% on ARC-AGI-3’s Semi-Private set with the Standard harness and 99.9% using a Provider Adapter that preserves opaque reasoning state between requests. With the Provider Adapter, Astra used fewer actions than the median human tester on 96% of levels. The human baseline drew on about 500 members of the general public. Astra’s API rates are $10 per million input tokens and $50 per million output tokens. GPT-6.1 Sol’s [lower price]() is $2 per million input tokens and $10 per million output tokens, one-fifth of Astra’s rate in each category. Third Moment Research co-founder Jeffrey Yan said, “GPT-6.1 Sol is currently our most-used model given its token efficiency. It completes tasks with the fewest tokens while delivering performance similar to GPT-6 Astra.” These measurements address distinct settings: a finance contest, computer-use benchmark tasks and simulated durations. They do not establish that Astra completes a broad range of workplace tasks faster, nor do the research applications demonstrate trading results. Third Moment Research’s system uses several models on extended finance workflows, while Astra’s computer-use scores come from separate benchmark evaluations. ## Links in this article - [lower price](https://www.tokenpost.com/news/technology/25466)