nesformer: a working NES CPU compiled into Qwen3.5 weights
A compiler, not training, turns a stock Qwen3.5-4B checkpoint into a working NES 6502 CPU: the model predicts the processor's execution log one token at a time and runs games like Super Mario Bros., very slowly.
Why it matters
Weights as a program: exact, deterministic computation compiled into a standard model checkpoint that loads with the normal Qwen class and no custom code.
What you could build with it
A computer-science lecturer could use this as a hands-on lab: students point the compiled-weights CPU at a small ROM, watch each instruction emerge as tokens, and compare the trace against a conventional emulator to learn how processors and language-model state really work.
Does it hold up?
A working demo, not a practical emulator: it plays Super Mario Bros. at 1-6 seconds per frame, so its value is as a teaching artifact and a proof that compiled weights can do exact computation.
Built with nesformer
- Simon Willison flags nesformerSimon Willison's Weblog · Notes that Drew Breunig baked a working NES 6502 CPU into Qwen3.5-4B weights, with the Opus 5.5 prompts used to build it shared in an HN comment.
- Show HN: A working Nintendo in ~1GB of Qwen weightsHacker News · 292-point HN thread where the author explains the build, the conformance tests used, and that the system runs 150-290x slower than a real console.
- Nesformer compiles an NES CPU into Qwen weightsLockerGnome · Explainer covering the 151 opcodes in the weights, the 3.3 MB compressed checkpoint, and the host program that draws the picture and reads the controller.
Learn more
First spotted on github: source.
More AI models
VoiceStudio: open-source, fully-local ElevenLabs alternative for voice cloning, dubbing, and audiobooksdebpalash/VoiceStudio is a desktop app that runs voice cloning, voice design, video dubbing, dictation, transcription…models · JEV 0.75Tencent Hunyuan Hy3: open 295B MoE with 21B active under Apache 2.0Tencent open-sourced Hy3, a 295B-parameter mixture-of-experts model with only 21B parameters active per token, under…models · JEV 0.73NVIDIA Isaac Lab 3.0 GA + Isaac OS open runtime + Agile ONE SN humanoidMajor Isaac platform expansion at GTC Berlin Oct 21: Isaac Lab 3.0 is generally available as the open-source robot…models · JEV 0.73NVIDIA Nemotron 3.5 Retriever collection: Embed 8B/1B, Rerank 8B, Parse 631MNVIDIA's Oct 21 collection drop for agentic retrieval: Nemotron-3.5-Embed in 8B BF16, 1B BF16 and 1B NVFP4…models · JEV 0.72Kimi K3.1 Pro: Moonshot's 2.8T open-weight flagship officially launchesMoonshot AI officially launched Kimi K3.1 Pro on Oct 23, a mixture-of-experts model with 2.8 trillion total / 104…models · JEV 0.72NVIDIA Cosmos 3.5 world models: now openly availableNVIDIA's Cosmos 3.5 family of open world foundation models for robots, autonomous vehicles and vision agents is now…models · JEV 0.71
Get the week's best AI launches, plus 3 ideas worth building
One email every Saturday. Ranked by traction, not hype. Free.