GPT-5.6 Sol Completes Factorio Peaceful Mode in 13.5 Hours

AI Agent Completes Factorio Run

GPT-5.6 Sol, operating within Codex with a low thinking level, completed a peaceful mode Factorio run. The AI agent finished the game, which involves building and managing an industrial factory, in 13.5 hours of in-game time. This marks a notable step in AI's capability to handle complex, goal-oriented tasks within a simulated environment.

Interaction Method and Constraints

The AI interacted with Factorio using a Python-based game manager and a small library. This setup allowed Sol to play programmatically while adhering to the same limitations faced by human players using the graphical user interface, such as player reach and fog of war. This method ensures the AI operates under realistic game conditions.

Performance and Challenges

During the run, Sol demonstrated unexpected effectiveness in early stages but encountered significant scaling issues in the later game, particularly after blue science research. The final rocket launch required Sol to manually ferry supplies to compensate for its inability to connect certain production systems efficiently. This highlights both the AI's problem-solving capacity and its current limitations in managing large-scale, interconnected systems.

Improvements Over Previous Models

This achievement represents an improvement over prior attempts with GPT-5.5, which failed to utilize transport belts, a fundamental game mechanic in Factorio. The successful completion by GPT-5.6 Sol indicates advancements in the model's ability to understand and apply complex game logic and mechanics.

Implications for AI Research

Autonomous Factorio gameplay is viewed as a valuable medium for understanding AI model capabilities. The game's analogy to programming, combined with its visual legibility, makes it a suitable testbed for observing and discussing AI behaviors. Future plans include testing other models, different thinking levels, and non-peaceful game modes.