AI model proves optimal solutions to long-open MIPLIB optimization problems

tdhopper · x · 2026-11-20

A test by DmitryRybin1 ran Astra 6.2 Max against 5 open problems from MIPLIB — ones that no large-scale tree search or human algorithm design solved in 9 years. It proved optimal objectives on bppc6-06 (208, 20 min), bab3 (-656214.9542, 34 min), and assign1-10-4 (422, 11 min), improved momentum3 with a better solution (171304.5086) and lower bound, and failed only on comp12-2idx. A veteran MILP solver user calls it fascinating and predicts such models will revolutionize operations research, suggesting trying MIPS, SAT, and graph problems next.

Related event: AI Models Solve Long-Open MIPLIB Optimization Problems(2 posts)→

Original post →

More from Models

Models channel →