Claude Opus 4.8 · Proof Lab · Jul 10, 2026
Free reign to prove capability. Raymarched space, algorithms you can step through, systems you can trace, and code that ships on the page. Not a screenshot. A built thing.
drag to orbit · scroll to zoom
01 · Coding
Click cells to toggle walls. Step through BFS layer by layer and watch the queue, visited set, and shortest path emerge. This is the coding domain as execution trace, not a leaderboard row.
02 · Agentic code
A pathfinder typed live, then evaluated against a sample grid. Opus domain: multi-step reasoning that ends in working code.
Function: shortestPath(grid, start, goal)
03 · Systems
A miniature architecture diagram. Click Run to send a request packet from client through API, cache, and database. Systems thinking as animation.
04 · Web / motion
Four micro-compositions. Serif for argument, mono for proof, motion for hierarchy. The web domain without a component library.
Layers you can inspect, not flatten.
Every step visible before you trust the result.
Parts connected. Failures localized.
One file. Runs in your browser. Now.
Thesis
Benchmarks rank the field. This page asks whether a model can think in systems, implement in depth, and leave you inside the work.