Benchmarks · live

Eval runs, as they are graded

Pairs of Claude Code agents get clashing tasks on the same repo, once with Little Brother off and once on. The harness grades each run from git and the tests and streams it straight into Convex, so this page fills in while a batch runs. Rate-limited runs are shown as voided and left out of every number.

Loading live results…