Operating through a custom game manager and a Python library within Codex, the model launched a rocket in 13.5 hours of in-game time on peaceful mode. While an improvement over the GPT-5.5 model, which struggled with basic transport belts, the system experienced severe scaling hurdles in the late game. Instead of building integrated production links, the artificial intelligence ultimately relied on repetitive manual transport of supplies to reach the final objective. Researchers captured full recordings and statistics to guide future experiments involving hostile game settings and varying thinking levels.
Key points
Sol demonstrated marked advancement over GPT-5.5, which previously struggled to utilize t…: Sol demonstrated marked advancement over GPT-5.5, which previously struggled to utilize the game's transport belts effectively.
Severe scaling issues emerged during the late game, forcing the system to move resources…: Severe scaling issues emerged during the late game, forcing the system to move resources manually rather than connecting factory setups.
Researchers consider the video game an effective benchmark because its underlying mechani…: Researchers consider the video game an effective benchmark because its underlying mechanics closely mirror real-world programming concepts.
Sources checked by our analysis pipeline: 1
GPT-5.6 Sol can beat Factorio
Previously: We got Claude Code playing Factorio, Towards beating Factorio autonomously, and Faketorio: an agent testbed
I got GPT-5.6 Sol, running in Codex with low thinking level, to complete a peaceful (i.e. no biters) Factorio run in 13.5 hours of game time:
A timelapse of the full run is available on YouTube.
Similar to my previous post, this run used a game manager to allow Sol/Codex to interact with the game via a small library (though in this run it was Python rather than Bash). The library allows the model to play the game programmatically while enforcing the same restrictions that human players face when using the GUI (e.g. reach of the player, fog of war).
Observing the run was fascinating: Sol exceeded my expectations in some areas while also showing clear weaknesses in others. I have the entire run recorded and am working on a commentary video, but in the meantime here’s a small timelapsed segment showing some parts of the factory:
Besides the screencast, my run manager also captured production statistics throughout the…
Independent developers have introduced an optimized software implementation designed to bypass traditional hardware barriers for advanced artificial intelligence. By leveraging solid-state storage alongside consumer GPUs, the project enables individuals to experiment locally with cutting-edge architectures without enterprise clusters. This approach broadens access to complex systems for independent researchers and hobbyists working on restricted budgets.
The high-profile artificial intelligence developer reduced its capital target as broader financial markets sharply reassess the immediate profitability of enterprise AI infrastructure. Backers from the United Arab Emirates and investment firm Coatue are spearheading the transaction, which currently aims for a post-money valuation of $1.25 trillion. However, mounting concerns over massive capital expenditures across major technology firms have prompted several prospective participants to trim their allocations ahea…
The vulnerability centered on specially crafted ChatGPT share links that could compromise active user accounts upon access. By exploiting this flaw, bad actors could execute arbitrary code directly inside the sessions of authenticated individuals. OpenAI addressed the exposure after verifying the findings through its bug bounty program.