GPT-5.6 Sol Completes Peaceful Factorio Run

Ryan Madden said on Oct. 31, 2026, that OpenAI’s GPT-5.6 Sol, running through Codex at a low thinking level, completed a peaceful Factorio run in 13.5 hours of game time, marking another step toward autonomous long-horizon model play. Madden used a game manager and Python library to let Sol interact with Factorio programmatically while preserving human-like constraints such as limited player reach and fog of war. The run showed both progress and persistent weaknesses. Sol initially progressed through early factory automation and science research more effectively than earlier attempts, prompting Madden to extend the run beyond his original blue science milestone. Production statistics and annotated research milestones captured by the run manager suggested strong early gains. Later, however, scaling problems emerged in the late game, when Sol failed to connect some production systems and relied on repeated manual-style supply ferries to advance the factory. The rocket launch was eventually completed, but it exposed inefficiencies in Sol’s long-term planning and spatial coordination. Madden highlighted the result as a meaningful improvement over prior GPT-5.5 attempts, which struggled even with basic transport belt usage. He also said the run offers a useful testbed for understanding model capabilities, noting that Factorio resembles programming tasks while remaining visually legible to non-experts. Madden plans to release additional commentary, test other models and reasoning settings, and explore more challenging non-peaceful runs.