AgentBench vs PyRIT
AgentBench
github.comPyRIT
microsoft.github.ioSide by side
| Figure | AgentBench | PyRIT |
|---|---|---|
| Price from | — | Free |
| Free plan | — | Yes |
| Free trial | — | — |
| Open source | — | — |
| Web | — | Yes |
| Linux | Yes | Yes |
| API | Yes | Yes |
| Self-hosted | Yes | Yes |
| Plans published | 0 | 1 |
| Spec score · LLM Evaluation Tools | 7.8 (#10) | 7.5 (#18) |
| License | The repository is licensed under Apache License 2.0. | The GitHub repository identifies PyRIT as MIT licensed. |
| Purpose | AgentBench is a benchmark for evaluating the reasoning and decision-making abilities of LLMs acting as agents. | PyRIT is an open-source framework for security professionals and engineers to identify risks in generative AI systems. |
| Support | The project invites users to join its Slack for questions and collaboration, and lists a Google Group for questions about AgentBench FC. | The project uses GitHub Issues for bugs, feature requests, and questions, and says support is limited to the resources it lists. |
Plans
In their own words
PyRIT
PyRIT is a free, MIT-licensed framework for security professionals and engineers assessing risks in generative AI systems. It supports single-turn and multi-turn red-team strategies such as Crescendo, TAP, and Skeleton Key. Scenario evaluations can be repeated across hundreds of objectives, including content harms, psychosocial risks, and data leakage. CoPyRIT provides a web interface for interacting with AI systems, recording findings, and collaborating with a team. Documented targets include OpenAI, Azure, Anthropic, Google, HuggingFace, custom HTTP endpoints and WebSockets, and web apps tested with Playwright. PyRIT stores conversations, scores, and attack results in SQLite or Azure SQL and can export results. Scoring supports true/false, Likert, classification, and custom scorers, including LLM-based and Azure AI Content Safety methods. Scanner, GUI, Framework, pyrit_scan, and pyrit_shell modes are documented. Installation is available through Docker, pip, or uv, with Python 3.13 recommended locally.