Third Moment Research — research organization

Overview

Third Moment Research builds agents that improve through analysis of their own behavior and synthetic scenarios targeting recurring weaknesses. The evidence here is OpenAI's customer report, not an independent evaluation. (source)

Key People

• Adam Khakhar and Jeffrey Yan are identified as co-founders. (source)

Models & Products

• The research system uses Codex with Astra, GPT-6 Sol and GPT-6.1 Sol. An outer harness examines unfinished trajectories and chooses retry, termination or human review. (source)

Recent Activity

• 2026-10-14: OpenAI reports 10,000 concurrent agents on investigations lasting 36-48 hours. Quantitative-research agents use a Dockerized environment with a terabyte of data, documentation and a research repository. (source)
• Codex identified caching and memory-packing changes that reduced one backtest from approximately 3 hours to 3 to 5 minutes, reducing polling and context pollution. This is one workflow optimization, not a general speedup. (source)

Strategic Position

Interpretation: the report connects sustained research to environment efficiency and explicit continuation decisions. Daybreak tests simulation environments for escapes before launch; no escape-prevention rate is published. Hundreds of thousands of collaborating subagents remain a roadmap goal. (source)

Related

• Agents (LLM Agents) · Agent Runtime Containment · OpenAI — orchestration and security boundaries for autonomous research. (source)