Jev beats Opus 5.5 and GPT-6 Astra: Rubik's Cube race
A randomly scrambled cube, solved the way people solve it.
On every API call the model sees the cube and picks its next step: which piece and which algorithm.
Results:
Jev: solved in 9.5s, 78 moves, 1 wrong pick
GPT-6 Astra: solved in 1m 27s, 94 moves
Opus 5.5: solved in 1m 47s, 102 moves
An LLM isn't the right tool for every job. Classifiers and LLMs work best together, and we're going to keep testing where that line goes. Next up: Jev as a judge in the real world.
Someone just moved from the weekly plan of my golf app to the yearly one 🚀
They started on September 10 and paid $4.99 a week three times. The yearly plan costs $39.99, the price of eight weeks. Apple switches them over when the week they paid for ends, so the money arrives in six days.
I do not talk to customers, so this is how they talk to me.
@bimbobinm You sound angry. The point isnt to let llm solve cubes. Its to showcase that not everything needs generative AI. A cube is just something I wanted to try because me and the Missy's are learning to solve cubes lol