UrBenchUr Solutions

The same task, every new model.

Each tile is one prompt. Inside are the results, one per model, exactly as shipped.

Craft

A Minecraft clone with the Cube World look, from one underspecified prompt.

26 runssince 2026-05Google GeminiAnthropicOpenAIOpenCodeXZ.ai

Star Pond

A gravity toy for kids, from a one-line prompt.

6 runssince 2026-08OpenAIAnthropicX