TenaBench
An open-source benchmark that tests whether AI models can work as real software engineering agents.
- Status
- active
- Built with
- TypeScript · Node.js · Express 4
- Source code
- github.com/Saba-Burduli/tenabench
What it is
A “Real-World Agent Capability Lab”: TenaBench puts an AI agent inside a real Git repository and gives it realistic engineering tasks. The agent has to decide what to inspect, how to implement and test, and when it’s done.
What it measures
Repository understanding, code search, feature implementation, debugging, testing, refactoring, Git workflows, code review, research, failure recovery and long-context reasoning.