Introducing SWE-2: Pushing the Pareto Frontier

Open paper
Benchmark Model Rank Results
code-generation-on-deepsweSWE-2 (mini-swe-agent)–Pass@1: 0.73
code-generation-on-terminal-bench-2-1SWE-2 (devin)–Accuracy: 92.8
code-generation-on-terminal-bench-4-0SWE-2–Accuracy: 27.3

Rank counts only results with a code link.