FrontierSWE

Measures: An agent gets up to 20 hours per task on 34 ultra-long-horizon problems (from-scratch implementations, performance optimisation, scientific computing, visual reasoning, AI research); each task is graded with partial credit and the score is the mean over 5 trials (mean@5).

Maintainer: Proximal Labs
Introduced: 2026
Status: Active
Saturation: Headroom
Typical provenance: Independent
In scoring set: Yes

Official leaderboard

Leaderboard

Show independent results only

Every result we hold, whoever ran it. Only independent and third-party rows count towards a score.

No results to show.