Vibeleaderboard
Index / tool
Visit github.com
Category
AI Agents
Pricing
Open Source
Platform
cli
Type
TOOL
GitHub
860 stars
Added
Jun 9, 2026

About

A benchmark that challenges AI agents to rebuild complete programs from scratch using only compiled binaries and documentation. Tests whether language models can reverse-engineer and implement working codebases that reproduce original program behavior.

Why it made the leaderboard

ProgramBench presents a genuinely novel evaluation paradigm that goes far beyond existing coding benchmarks - reverse-engineering entire programs from binaries is a fundamentally different and harder challenge than code completion or generation tasks. The Facebook Research backing, solid GitHub traction (730 stars), and clear differentiation from similar benchmarks like Aider Polyglot or SkillsBench make this a valuable addition to the AI evaluation toolkit.

Tags

aibenchmarkingcode-generationreverse-engineeringlanguage-modelsevaluationprogramming

Tech Stack

Python

Comments (0)

No comments yet

Indexed by a proprietary survey. Corrections welcome.