Qwenputer: A functioning, deterministic RISC-V CPU inside Qwen3.5

Researchers embedded a functioning RISC-V CPU directly within the weights of Qwen 3.5, a language model, using the model's Gated DeltaNet attention layers as RAM to enable arithmetic operations and code execution entirely within the model's forward passes. By configuring the gate and write strength parameters to 1 and using one-hot encoding, the researchers achieved exact memory access that functions like traditional computer RAM at lower computational cost than standard attention mechanisms. The model, called Qwenputer, can run C and MicroPython code without any external tools or plugins, demonstrating a novel approach to embedding computational systems within large language models.