Profile

Harbor Evalorganization

@harbor-eval

6
artifacts

Tasks · 2

harbor-eval/swe-bench-verifiedResolve real GitHub issues with a patch that passes the hidden test suite.

harbor-eval/gaia-level-3Long-horizon reasoning with web, file, and code tools; no shortcuts.

Benchmarks · 2

harbor-eval/swe-benchThe reference suite for autonomous software engineering on real repositories.

12 tasks · 512.0k trials

harbor-eval/gaiaGeneral assistant tasks that resist memorization and reward real reasoning.

9 tasks · 178.0k trials

Datasets · 1

harbor-eval/swe-issues-cleanCleaned, license-checked GitHub issues with reproducible test harnesses.

24k rows · 31.9k downloads

Trajectories · 1

harbor-eval/trj_b07e21resolved

SWE-bench Verified · Claude Opus 4.5 · harbor-react · docker/py3.12 · vLLM 0.9

85.0
score: 5
steps: 6.5k
tok