GPT-5.6 Sol can beat Factorio on peaceful mode

Hacker News by 2 min read 25x views
GPT-5.6 Sol can beat Factorio on peaceful mode

Share Post

Oct 31, 2026

Previously: We got Claude Code playing Factorio, Towards beating Factorio autonomously, and Faketorio: an delegate testbed

I got GPT-5.6 Sol, operating in Codex alongside low thinking level, to complete a peaceful (i.e. no biters) Factorio run in 13.5 hours of equivalent time:

A timelapse of the complete run is accessible on YouTube.

Similar to my previous post, this run used a equivalent director to authorize Sol/Codex to engage alongside the equivalent via a small archive (though in this run it was Python fairly than Bash). The archive allows the example to perform the equivalent programmatically during enforcing the identical restrictions that individual players visage whenever using the GUI (e.g. attain of the player, fog of war).

Observing the run was fascinating: Sol exceeded my expectations in several areas during additionally showing apparent weaknesses in others. I have the complete run recorded and am operating on a commentary video, but in the meantime here’s a small timelapsed section showing several parts of the factory:

Besides the screencast, my run director additionally captured manufacturing data throughout the game. Here’s a diagram of total asset production, annotated alongside discipline investigation milestones:

I had planned to terminate the run following blue discipline research, but Sol’s amazing effectiveness convinced me to keep it going. Unfortunately, the scaling issues I idea would happen in the blue discipline phase appeared afterward and slowed Sol’s related advancement significantly.

The missile initiate was lone achieved through important ‘manual’ attempt on Sol’s part as it papered complete its late-game failures to nexus certain manufacturing systems by often ferrying supplies between areas of the factory.

Still, I was pleased alongside the betterment complete former attempts alongside GPT-5.5 which suffered from the model’s failures to use transport belts at all.

I’m optimistic concerning autonomous Factorio perform as a average for construction intuition concerning the nature of example capabilities. In many ways playing Factorio is akin to programming but alongside the advantage of being much additional legible to informal observers, making it convenient for grounding conversation of example behaviors.

I appearance onward to evaluation another models and thinking levels, as fine as non-peaceful runs. Stay tuned!

Other Article Hacker News
↑
Close Right Ads
Close Left Ads