A full RISC-V CPU implemented inside Qwen3.5-4B's weights lifts math accuracy from 48.6% to 100%
Dan_Jeffries1 · x · 2026-10-29
A demo shows a full RISC-V CPU implemented within Qwen3.5-4B's weights: instead of guessing math answers, the model executes instructions internally, using its Gated DeltaNet state as memory.
- Arithmetic accuracy jumps from 48.6% to 100%
- The model executes code internally and scores 97.5-100% on tasks where the base Qwen struggles
- On DeepMind math and long-integer tasks it matches or beats the base model with Python tool use
Takeaway: a CPU inside a language model is real — computation happens in the weights, not via external tools.
Related event: Qwenputer compiles a 32-bit RISC-V CPU into Qwen3.5-4B weights(8 posts)→
More from Models
- $5 in evals cut his AI bill 74%: Claude Haiku 5.5 beat Gemini 3.8 Flash at 1/4 the price — MicahBerkley · 2026-10-29
- Gemini silently swaps in another model when 'Pro is in high demand' — General_Internet_511 · 2026-10-29
- Meta's Alexandr Wang praises Muse's terse, task-first answers as 'anti-slop-text' — alexandr_wang · 2026-10-29
- Weight forensics with Claude: Qwen3.5-27B appears grown from 9B via layer looping — CatAstro_Piyush · 2026-10-29
- OpenAI's Astra becomes main target for Chinese AI labs after Anthropic tightens defenses — The Decoder · 2026-10-29
- Openly available GLM model completes multi-stage cyberattacks in controlled tests — S_OhEigeartaigh · 2026-10-29