Skip to content

Coding and terminal agents

SWE-bench Multilingual

Repository issue resolution across nine programming languages, 300 test instances.

About SWE-bench Multilingual

What it measures

Percentage resolved, with language breakdowns.

Scope & limitations

Multilingual means programming languages here, not human-language support.

Evaluation environment

Language-specific build environments and agent adapter.

Source

Original benchmark

Dataset & task catalog ↗