← Benchmarks

Cursor Grok 4.5 High Fast · Proof Lab · Jul 9, 2026

Not a scoreboard.
A demonstration.

Free reign to prove capability. One page. Live physics, generative form, typed code that runs, and motion that responds to you. No screenshots of someone else's leaderboard. The artifact is the argument.

canvas physics procedural art live JS spring motion

move the cursor · the field follows

It writes. Then it runs.

A small agentic loop typed live: parse, plan, execute. Hit Run to evaluate the finished function against sample inputs. This is the coding domain as behavior, not a bar chart.

live buffertyping
stdouteval

Function under test: rankModels(scores) returns domain winners from a sparse score map.

Waiting for Run…
Chars
0
Lines
0
ms type
0

Grab a body. Break the orbit.

Soft-body-ish particles with springs, gravity, and mouse drag. Not a video of physics. Actual integration on your machine, sixty times a second.

verlet · springs · dragclick + drag a node
Nodes
0
Springs
0
Energy
0

A field that never repeats.

Procedural flow field drawn every frame. Seed, curl, and density are yours. This is generative stills without calling an image API: math as medium.

flow field · simplex-ish noise— fps
seed42
curl35
density55

Interface as choreography.

Four micro-compositions. Hover them. The page is the product: typography, timing, and restraint. This is the web-animation domain without a framework.

Latency

Response under a frame. Feel before you measure.

Hierarchy

Serif for thought. Mono for proof. Quiet chrome.

Continuity

One scroll. Four domains. No dashboard clutter.

Agency

You drag, twist, run. The model leaves the controls with you.

Benchmarks tell you who leads a table. This page tells you whether a model can ship an experience under free reign: taste, systems, and motion in one sitting.