GPT-5.6 Sol can beat Factorio on peaceful mode

GPT-5.6 Sol managed to complete a peaceful Factorio run in 13.5 hours of game time, a feat that surpassed expectations and highlighted the model's growing capabilities. The agent, running in Codex with a low thinking level, demonstrated surprising effectiveness during the early stages of the game, particularly in resource production and science research milestones.

However, as the game progressed into the blue science stage, scaling issues emerged, slowing Sol's progress and significantly impacting its overall performance. The model's inability to utilize transport belts in certain areas of the factory led to a series of workarounds, including manual intervention to ferry supplies between production systems and connect key elements.

Despite these challenges, the success of the rocket launch, achieved through substantial manual effort, showcased Sol's ability to navigate complex production networks. This experience underscores the potential of autonomous Factorio play as a valuable tool for understanding model capabilities and making the complex interactions within the game more accessible to observers.

Moving forward, the author looks forward to testing other models and conducting non-peaceful runs, promising further insights into the ever-evolving landscape of AI in gaming.