An AI computing enthusiast has demonstrated Doom running on a custom CPU designed by GPT-5.6 Sol. Angel (@Angaisb_) shared a technical and visual feast with a video of classic Doom running within a pulsing schematic of the CPU, in the sandbox provided by the Turing Complete game environment. Clearly, this shows that GPT-5.6 Sol is capable of constructing a virtual computer system from the basics, and one that is accurate enough to boot and play a port of the original Doom. The model dubbed its CPU design ‘Codex-R32.’
Angel knows Doom is a hugely popular demo to show computing capabilities, despite it being quite a vintage nowadays (1993). So they must have thought it a good idea to show how far agentic coding has come with this tried and trusted ‘people asked for Doom’ metric. They were impressed by the results, concluding that it is “truly insane that we can do stuff like this.”
Despite the accomplishment, another rule of the internet is that there’s always someone who thinks you haven’t pushed far enough. So there were the inevitable remarks about having a shot at Crysis next. Likely grasping for a witty comeback, Angel asked their AI to whip up a response to the unsolicited stretch goal. It didn’t disappoint, suggesting they answer “Absolutely – I just need to build a GPU, add a few gigabytes of RAM, and make a schematic visible from orbit first.”
Indeed, Doom is a relative minnow compared to Crysis, though it still stretches the model and its performance in the sandbox is not what anyone would call playable. “The CPU is built from primitive logic components inside Turing Complete,” the Codex told Angel. “Doom uses the C-based PureDOOM port, compiled into native RV32IM machine code that runs directly on the custom Codex‑R32 CPU.”
Watching the video, you can see the live schematic of the custom CPU with a pulsing visualization of processor gates, registers, memory, ALUs, and other blocks within the Turing Complete computer-science education and puzzle game’s anything-goes sandbox mode. Live stats show the cycle counter, sim speed, memory/register values, and more.
Overall, with this demo, we have a neat visual demo of LLMs not just creating software, but also reinventing hardware at a capable enough level to pass the 'can it run Doom' test.
