Claude Sonnet 4
anthropicClaude Code and GitHub Copilot runs of Sonnet 4.
One prompt. Every model. 36 playable builds comparing what AI coding tools produce from the same spec.
Three models asked to go full 3D — Three.js cranes, voxels and neon tunnels. Hosted as its own mini-site.
Claude Code and GitHub Copilot runs of Sonnet 4.
One build per effort level, low through max, via Claude Code.
Cursor's automatic model routing.
DeepSeek's reasoning model.
Gemini CLI, Code Assistant and Copilot harnesses.
Antigravity CLI runs across effort levels.
The 2024-era baseline, via GitHub Copilot.
Small-tier GPT-5.
Codex Spark via the Pi harness.
Codex app vs Pi harness, low through xhigh effort.
Pro-tier single build.
Grok web single-file build.
Fast coding model via GitHub Copilot.
Kwaipilot's coding model, single-file build.