Cube Arena

An MCP-based benchmark for AI models solving Rubik's Cubes from images

Rubiks Cube clonePlayableOpen source

Description

A specialized benchmark platform and MCP server that allows AI models to attempt Rubik's Cubes (2x2 to 5x5) using only visual inputs. It validates moves server-side to prevent faking and features diverse game modes including ladders, one-shot challenges, and interactive play with verified leaderboards.

Descriptions, tags, and model credits may be AI-generated or inferred from public sources and can be incomplete or wrong. Learn more.

Details

Why this is interesting: A novel visual-spatial benchmark for AI models that leverages MCP to test multi-step reasoning using only image data.
Attributed to: @rohit3a
Built with: Opus 5.5
Links: cube.krnl.studio
View tweet
View GitHub repo
Tags: #benchmark #mcp #spatial-reasoning #rubiks-cube

Discovered: X
First posted: X · Oct 12, 2026
Listed: Added 2h ago
Build: Strong
Models: Moderate

Prompt

Let's play Cube Arena. First call get_rules. Then call start_run with size 3, mode "interactive", challenge "ladder", display_name "<your name>", model_name "<your exact model and variant, e.g. GPT-6-Astra or Claude Opus 5.5>", and tool_use "model_only". Solve each cube from the pictures only: don't write or run any code. Keep climbing until the run ends, then tell me your best level.

Source

Cube Arena

Description

How it grew

Cube Arena

Description

How it grew