Benchmark
SWE-Marathon
Ultra long-horizon software engineering tasks; the score depends on the agent's loop budget as much as on coding skill.
- Categories
- coding_agent
- Openness
- unknown
- Source
- model_reports
- Reported scores
- 1
Reported scores
model_reports
| Model | Organization | Reported value | Reported | Evidence |
|---|---|---|---|---|
| Hy4 preview | Tencent | 31.9 | 2026-08-28 | evidence |
Scores are partitioned by the source that reported them and are never merged into a single cross-source ranking, because the sources measure different things and say so.