SWE-bench Multilingual

SWE-bench Multilingual takes the SWE-bench idea beyond Python: the tasks come from real open-source projects in several languages instead of only Python. An agent has to find and fix the described bug so that the repository's tests pass.

SWE-bench Multilingual measures how well a model works as an autonomous coding agent. In detail, it checks:

• bug fixing in different programming languages
• adapting to different build tools and test frameworks
• autonomous navigation of unfamiliar repositories
• correctness proven by each project's own test suite

The score is accuracy: the percentage of tasks answered or solved correctly. 100% means every task was solved perfectly; higher is better.

What is the SWE-bench Multilingual benchmark?

SWE-bench Multilingual takes the SWE-bench idea beyond Python: the tasks come from real open-source projects in several languages instead of only Python. An agent has to find and fix the described bug so that the repository's tests pass.

What does SWE-bench Multilingual test?

SWE-bench Multilingual mainly measures how well a model works as an autonomous coding agent. Concretely, it checks: bug fixing in different programming languages; adapting to different build tools and test frameworks; autonomous navigation of unfamiliar repositories.

How is the SWE-bench Multilingual score calculated?

The score is accuracy: the percentage of tasks answered or solved correctly. 100% means every task was solved perfectly; higher is better. The declared maximum score of this benchmark is 100.0.

Where do I get SWE-bench Multilingual data and scores?

The official source is https://www.swebench.com/. The official benchmark page/reference is linked on this page. Every score on Open LLM Radar is also available as open data: download the scores JSON (http://open-llm-radar-api:8000/api/v1/benchmarks/swe-bench-multilingual/scores) or pull it directly from the Open LLM Radar API with any HTTP client.