This interactive 3D viewer plots learned features for nine open transformer models — Gemma 4 E2B, Ministral 3B, MedGemma 1.5 4B, Qwen3 0.6B and 8B, Qwen3.6 35B MoE, Llama 3.1 8B, GPT-OSS 120B MoE, and Bonsai b1.58 2B — and shows the universal four-stage circuit they share: broadcast, domain, entity commitment, and prediction. Compare any model against the 1-bit Bonsai dissolution to see what happens when the circuit disappears. The viewer uses Three.js to render every feature point and a 12-second demo tour explains the four stages. Vindex IDs link back to the corresponding Hugging Face repositories so you can verify each model independently. Drag to orbit, scroll to zoom, click a feature for a layer-by-layer trace.