Introducing the World's Strongest Sokoban Solver
Not investment advice. Muhabbit scores how much a story may matter, not which way to trade.
Soksolv, a new Sokoban solver built by an independent developer using Anthropic's Claude Code, has solved more levels than any previously published solver on both standard benchmark suites, including 24 levels no automated system had solved before, and runs about 3.2 times faster in total time on levels other solvers can handle. The project was completed without handwritten code in a single 14-day Claude session using roughly 430 subagents at an API-equivalent cost of about $3,500, plus 15,000 vCPU-hours of AMD EPYC compute for research and benchmarking. On the large test suite it leads Festival by 10 levels at the 10-minute mark, and on the open test suite by 18 levels. The result matters as a public data point on agentic coding economics and as a step forward for a PSPACE-complete puzzle domain where small boards hide massive search spaces. What to watch next is whether the approach generalizes to other hard combinatorial search problems and whether published results from rival solvers are re-verified on comparable hardware.
Muhabbit's AI wrote this summary in its own words from Hacker News's report. It can get details wrong, so check the source.
Named in this story
Soksolv, Claude Code, Anthropic, Festival, Sokolution