AI Model Embeds RISC-V CPU to Run DOOM Inside Neural Network Weights

AI Model Embeds 32-bit RISC-V CPU to Run DOOM Inside Its Weights

5 4B language model's weights, enabling it to run the game DOOM within the model itself.

Edited by The Intel Brief Desk · 3 Nov 2026, 00:56 · 1 min read

Key takeaways

5 4B large language model (LLM), creating a hybrid system called "Qwenputer."
This embedded CPU within the model was able to run the classic game DOOM, demonstrating functional code execution inside the neural network.
The resulting hybrid, Qwenputer, can execute programs such as DOOM on the built-in CPU embedded in the AI model.

What we know

5 4B large language model (LLM), creating a hybrid system called "Qwenputer." This embedded CPU within the model was able to run the classic game DOOM, demonstrating functional code execution inside the neural network. 5 4B LLM’s weights. The resulting hybrid, Qwenputer, can execute programs such as DOOM on the built-in CPU embedded in the AI model.

Why it matters

This experiment shows a novel way to combine traditional computing hardware concepts with neural network architectures by embedding a CPU directly into an LLM’s parameters. It highlights the potential for AI models to incorporate programmable elements, which could open new directions for AI design and functionality beyond pure language tasks. The project serves as a proof of concept for running executable code inside a language model.

What is still unknown

No independent tests of the hybrid model’s accuracy or performance have been reported. Details about supported programming languages, broader software compatibility, or plans for public release remain unclear.

Get the daily brief by email.

AI Model Embeds 32-bit RISC-V CPU to Run DOOM Inside Its Weights

What we know

Why it matters

What is still unknown

Sources

Related