The Qwenputer Project
AI engineer Divyansh Agrawal has revealed the Qwenputer. Described in its most basic terms, Qwenputer is an experimental blend of the stock Qwen 3.5 4B AI LLM with a 32-bit RISC-V CPU compiled and inserted directly into its weights. In use, Qwenputer can run processor arithmetic operations – or even run C and MicroPython code on its built-in CPU, eschewing any external runtime or tool use. Moreover, it can run Doom, albeit very slowly.
QWenputer by @divagr1925 is so cool! but... can it run DOOM? Ofc yes! E1M1 on the RISC-V CPU compiled into Qwen3.5-4B weights. will opensource too https://t.co/hEa46rD6CJOctober 29, 2026
State as RAM and Compilation
The implementation of Qwenputer raises some interesting concepts and solutions. For example, the developer used Qwen’s Gated DeltaNet state as RAM, which is noted as a clever architectural trick. While using these fast weights as RAM is not an entirely new concept, modifying Qwen 3.5 in this manner appears to be a first.
Unlike traditional LLM creation, adding a CPU does not involve training. This RISC-V CPU implementation consists of transformer layers in front of Qwen's own, where a compiler computes every weight rather than relying on learned parameters. The built-in CPU passes official RISC-V tests and has executed millions of instructions free-running without error.
Execution and Filesystem
The model uses the embedded CPU to send instructions and read them back with no external tool or runtime calls. This provides exact calculations that overcome the arithmetic weaknesses common in many LLMs. Additionally, users can execute code directly on the internal CPU with C and MicroPython compatibility.
Resource Demands and Outlook
The developer also implemented a persistent filesystem utilizing the DeltaNet state. Storage includes slots holding multiple files that remain persistent if the state is saved at the end of a conversation.
Due to its expanded resource use, Qwenputer effectively doubles the size of the underlying Qwen 3.5 4B LLM to nearly 19GB. Consequently, testing the current version requires a workstation equipped with 24GB of VRAM or more, while ongoing efforts aim to optimize performance and coding speed.
The project remains experimental, serving primarily as a technical exploration of what is possible rather than a targeted commercial tool. Nevertheless, the creation of this custom LLM introduces novel concepts for hybrid computing architectures.
In addition to the above-linked blog, you can learn more about Qwenputer from its MIT-licensed code on GitHub. If you want to jump straight in and tinker with this fascinating project, you can now grab this LLM from Hugging Face.
