Lab

Same prompt, different models. Play every build.

playToken savers, testedClaude Opus 5.5 and Sonnet 5.5 · 2026-11-03Caveman, ponytail, rtk and context-mode, run against plain Claude Code. Only ponytail cut costs on nearly every job. Open its builds next to the plain ones.

playOne landing page, ten waysmultiple models · 2026-10-30Six AI models built a waitlist page for the same made-up startup, with no design skills installed. Open all ten pages.

playSquishy gummy bears, eight waysmultiple models · 2026-10-15Eight AI models got the same prompt: build squishy gummy bears in three.js. Play every build.

On YouTube

• ▶ Is Google back? Gemini 4 Argon against the other models ↗

playBloons TD, 41 waysmultiple models · 2026-10-1441 AI builds of one prompt: a Bloons-style tower defense game. Claude and GPT from low to max effort, Kimi K3.1 Pro, plus planner-and-coder teams. Play every build.

On YouTube

• ▶ The effort-level test: six models, every effort setting ↗
• ▶ Claude Haiku 5.5 tested, including its Bloons builds ↗