Program Bench

Program Bench is a benchmark for how well a model works as an autonomous coding agent. The official description says: "Evaluates code-generation agents by asking them to recreate a program's behavior from only a compiled binary and its documentation". Below we describe in simple words what benchmarks of this type generally examine.

Program Bench measures how well a model works as an autonomous coding agent. In detail, it checks:

• fixing real, open bugs from real software repositories
• planning multi-step changes, running commands and editing files by itself
• passing the project's test suite after its changes
• handling whole repositories, not just small isolated code snippets

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

Official methodology note: Evaluates code-generation agents by asking them to recreate a program's behavior from only a compiled binary and its documentation. It spans 200 tasks, from small CLI tools to large systems like FFmpeg and SQLite. Submissions are judged against over 248,000 fuzz-generated behavioral tests. In each task, the agent is given an executable and its documentation, but no source code, decompilation, or internet access. It must choose its own implementation language, build the full program from scratch, and pass a behavioral test suite comparing its output against the original binary.

What is the Program Bench benchmark?

Program Bench is a benchmark for how well a model works as an autonomous coding agent. The official description says: 'Evaluates code-generation agents by asking them to recreate a program's behavior from only a compiled binary and its documentation'. Below we describe in simple words what benchmarks of this type generally examine.

What does Program Bench test?

Program Bench mainly measures how well a model works as an autonomous coding agent. Concretely, it checks: fixing real, open bugs from real software repositories; planning multi-step changes, running commands and editing files by itself; passing the project's test suite after its changes.

How is the Program Bench 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 Program Bench data and scores?

The official source is https://programbench.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/program-bench/scores) or pull it directly from the Open LLM Radar API with any HTTP client.