Qwenputer Arrives: Packing the RISC-V CPU Instruction Set into a Model, AI Arithmetic No Longer Relies on External Tools
Developer Divyansh Agrawal released the experimental model Qwenputer on October 22. Built on Alibaba's Qwen3.5-4B, it implements a 32-bit RISC-V CPU and memory functions inside the model, enabling it to execute programs and perform precise calculations without calling external tools. In tests, large-integer multiplication accuracy rose from 55% to 97.5%, and the long-integer arithmetic benchmark NUPA reached 100%, compared with only 48.6% for the standard version. The cost is that the model size grows to about 18.8GB, each instruction requires generating multiple tokens and thus takes several seconds, and a dedicated driver is needed. The project has been open-sourced on Hugging Face and GitHub.
Why read this
AI score 66
The original text provides comparative figures for multiple arithmetic benchmarks along with the costs in size and speed, which can be used to judge the feasible boundaries of encoding an instruction set into model weights.
Topics
Body · AI translation
Qwenputer Launches: Packing the RISC-V CPU Instruction Set into a Model, AI Arithmetic No Longer Relies on External Tools
2026-10-30 14:12 IT Home - Guyuan
IT Home, October 30 news: Developer Divyansh Agrawal released the experimental model Qwenputer on October 22. Built on Alibaba's Qwen3.5-4B, it implements 32-bit RISC-V CPU and memory functionality inside the neural network, enabling the large language model to execute programs and perform precise calculations without calling external tools such as Python.
Citing a blog post, IT Home explains that traditional AI large models generate answers by predicting the next token, and are prone to errors when encountering large-integer multiplication and division or long-chain operations, usually requiring calls to an external calculator or code interpreter.
Qwenputer's approach is to "compile" the RISC-V RV32IM instruction set and related extensions into the model weights, and to use the internal state of Qwen3.5's gated DeltaNet as RAM, thereby building within the model a CPU capable of executing arithmetic, conditional branches, and programs.
Test results show that in large-integer multiplication, the original Qwen3.5-4B had an accuracy of 55%, while Qwenputer raised it to 97.5%; large-integer division rose from 72.5% to 97.5%; and tests involving multiple multiplications and divisions reached 100%.
On the long-integer arithmetic benchmark NUPA, Qwenputer reached 100%, while the standard version scored only 48.6%. The developer says the CPU has passed the official RISC-V tests, executing about 18 million consecutive instructions without errors.
The project also experimentally adds an "in-model file system" that can store up to 512 files, about 53,000 tokens, in the gated DeltaNet state and load them in subsequent conversations. In tests, answers using information saved from prior conversations achieved 90% accuracy, compared with only 30% for the control method that directly feeds prior conversation content into the model.
However, Qwenputer still has notable costs: the model size grows to about 18.8GB, nearly twice the original; each CPU instruction requires generating multiple tokens, so a calculation that originally took a few milliseconds may take several seconds; and running it requires a dedicated driver, so it cannot be loaded directly like an ordinary large model.
The project is open-sourced on Hugging Face and GitHub and can run on a single GPU with 24GB of VRAM.
Reference
Qwenputer: a fully functioning, deterministic RISC-V CPU inside Qwen3.5
Click here to visit divagr1925/Qwen3.5-4B-Qwenputer on Hugging Face
Click here to visit the GitHub project
Advertising statement: External redirect links contained in this article (including but not limited to hyperlinks, QR codes, passcodes, etc.) are used to convey more information and save selection time; the results are for reference only. All articles on IT Home containing external links include this statement.