banteg's Sokoban eval: community cuts 66-room LLM record from 3,819 to 3,063 keys in 24h
banteg · x · 2026-11-12
banteg launched "Sokoban Eval · Frontier," a 66-room Sokoban benchmark for language models where each room opens with the shortest route a published model run found; anyone can beat the record or prove a higher lower bound.
Within 24 hours the community compressed the total from 3,819 opening keys to 3,063, just 63 above the proven minimum of 3,000. Remaining unproven rooms: III-14 (5 keys to close), III-20 (10), III-21 (48). Submissions replay key-by-key in Jonah Ostroff's original PuzzleScript build; undo keys don't count toward length and every record replay is public.
More from Fun
- The 2026 bedtime joke: husbands are talking to something else now — hey_abusiddik · 2026-11-12
- AI-generated breakdance video wows Reddit with fluid motion — TradingCardGirl · 2026-11-12
- Claude scans 126,348 ICLR 2027 reviews: ~25K contain Markdown formatting — NielsRogge · 2026-11-12
- Bengio memes Jensen as the suffering triangle: startups, jobs and open-source all depend on him — NandoDF · 2026-11-12
- New AI metric joke: beyond adjusted profit, companies should report "Most Tokens Used" — MatthewChang · 2026-11-12
- CS professor's meme about 'that one co-author' resonates with academia — CSProfKGD · 2026-11-12