Four models solve the same Rubik's Cube: Microsoft 9B and Jev fastest, GPT-6 Sol stalls
Given the same scramble, four models each solve a Rubik's Cube in 12 moves, with Microsoft's 9B model and TypeSafe Jev fastest and cheapest while GPT-6 Sol is stuck on move 8.
A 9B model from Satya Nadella's Microsoft just beat Sam Altman's GPT-6 Sol at a Rubik's Cube, and it's built on Alibaba's Qwen Same scramble, same 12-move solution for all 4 models, 1 model call per move, real time Results: Microsoft Decision-1: solved in 5.9s for $151 per 1M solves TypeSafe Jev: solved in 6.8s for the same $151 Alibaba Qwen3.8 27B: solved in 10.4s for $5,040, open weights OpenAI GPT-6 Sol: stuck on move 8 when the clip ends, $31,200 per 1M solves OpenAI's flagship burns 206x more money than Microsoft's 9B model to make the same moves, and a free-to-download Qwen running at $0.15 per million tokens still finishes 4x faster than GPT-6 Sol
This post quotes another post: x.com/leopardracer/status/2108479372675977496