AI Proves Optimal Solutions to MIPLIB Open Problems Unsolved for 9 Years
IgorCarron · x · 2026-11-20
DmitryRybin1 tested whether Astra 6.2 Max could match 300 agents of Fable 5.2 on the Open Combinatorial Optimization dataset, tackling five MIPLIB open problems that resisted large-scale tree search and human algorithm design for over 9 years: bppc6-06 proven optimal at 208 in 20 min, bab3 optimal at -656214.9542 in 34 min, assign1-10-4 optimal at 422 in 11 min, momentum3 improved to 171304.5086 with a better lower bound of 135070.9986 in 150 min, while comp12-2idx saw no improvement. Long-time MILP practitioner IgorCarron calls it simply fascinating.
More from Models
- Naval-war game benchmark: author claims GPT 6.2 Astra crushes Opus 5.2 — RichardsonDx · 2026-11-20
- Qwen3.8-27B-pi-adapted GGUF quantization trends on Hugging Face — harness-hub · 2026-11-20
- Musk amplifies Grok Bot phone-call demo highlighting live transcripts — elonmusk · 2026-11-20
- Airbnb Cut Policy Classification Costs 99% With Gemma 4 E2B at 35ms Latency — petrusenko_max · 2026-11-20
- One neuron per language hits 99.1% language-ID accuracy, beating dedicated models — julianweisser · 2026-11-20
- 10 hands-on lessons on what actually happens inside an AI after you type a prompt — techartist_ · 2026-11-20