13 Models and 4 Agents on SWE Tasks: Go, Java, Python, Rust, TS
Summary
The SWE-rebench Leaderboard page presents the SWE-rebench v2 results for software engineering tasks, spanning 111 problems from 65 repositories within a May–July 2026 window. It lists model performance metrics (Resolved Rate, Pass@5, Cost per Problem, Tokens per Problem), highlights new model entrants, and provides links to the paper, data, and project discussions. The compilation emphasizes tool-assisted SWE tasks across multiple programming languages and environments.