abundant/swe-marathon
Coding
Ultra long-horizon software engineering tasks from SWE-Marathon.
Run this task
CLI:
inspect eval inspect_harbor/abundant_swe_marathon --model openai/gpt-5Python:
from inspect_ai import eval
from inspect_harbor import abundant_swe_marathon
eval(abundant_swe_marathon(), model="openai/gpt-5")Dataset information
| Harbor registry | abundant/swe-marathon |
| Inspect task | abundant_swe_marathon |
| Latest digest | sha256:862b01dc3c8d3bf5b8016ea68181c3963d62588ab03b928e5c6646cfce4e7b3f |
| Samples | 20 |
| Paper | arxiv |
| Source | https://github.com/abundant-ai/swe-marathon |
See Task Parameters for the parameter set shared across all Harbor tasks.