Evaluating DeepAgents CLI on Terminal Bench 2.0

DeepAgents CLI, a model-agnostic coding agent, was evaluated on the Terminal Bench 2.0 benchmark to measure its performance on real-world tasks. The benchmark includes 89 tasks across domains like software engineering and security, testing an agent's ability to operate in a terminal environment. To ensure clean, isolated, and scalable evaluations, the Harbor framework was used to run the agent in containerized environments. The DeepAgents CLI, powered by Claude Sonnet 4.5, achieved a score of approximately 42.5%, establishing a competitive baseline for future improvements. This process demonstrates a robust method for systematically testing and iterating on AI agents.