Abundant Labs

Run Harbor tasks in the cloud

Builder-made artifacts for rsi-bench-tasks that are too large for git (start-model adapters etc.); pinned by sha256 from task Dockerfiles. No benchmark held sets.

SWE-Marathon: an ultra long-horizon SWE benchmark

Turn a Lean 4 repository into a graded proof task: build it, cut a theorem's proof, and grade a solver's replacement.

Convert GitHub repos into long-horizon reverse engineer tasks

100 long-horizon reverse engineering tasks generated by ProgramSmith.