Someone ran strings against Claude Code and found a complete hidden leader and worker agent system already compiled inside the app behind two feature flags.
This guy vibe coded an app with Claude Code that scans your Rubik's Cube and shows you how to solve it in 20 moves or less.
There's a math fact called God's Number. No matter how badly a cube is scrambled, it can always be solved in 20 moves or fewer. This app proves it every
GPT-5.6 Sol vs Fable 5, same app, same prompt, very different outcomes.
→ Sol finished in less than half the time at the same reasoning level
→ Sol ignored the image generation request first try, then ignored the transparent background suggestion, justifying both on its own
→
Someone used GPT-5.6 Sol Ultra to build a Doom-like game where every single thing the game does runs as a SQL query.
Raycasting the scene, calculating RGB pixels, encoding them as colored terminal characters, handling movement, collision, enemy AI and combat. All of it SQL.
OpenAI's GPT-5.6 Sol can now run inside Claude Code and both setup options take minutes.
The official plugin adds /codex:review, /codex:adversarial-review and /codex:rescue straight to Claude Code so Claude writes and Sol critiques before you ship.
The proxy option is one
Running Fable 5 for everything is a waste of tokens. Here's how to use it only for what it's actually worth.
Install the Superpowers skill, then run one prompt that makes Fable ask what the project is for before reading anything, map the entire codebase, review across four
Someone asked GPT-5.6 Sol in Cursor to set up Blender MCP and render a realistic floating MacBook without ever opening Blender before.
Look at that result.