harithoppil/terminal-bench-2-trajectories

Terminal-Bench 2.0排行榜轨迹数据集包含从Terminal-Bench 2.0排行榜提交中提取的代理轨迹。每条记录包含提示（任务指令）、代理的响应和奖励（通过/失败）。数据集包含多个配置，如所有轨迹、仅通过的轨迹、ML/训练相关任务等。此外，数据集还记录了多个模型的表现数据，如Claude-Opus-4.6、Gemini-3.1-Pro-Preview等，以及ML任务的相关数据。

The Terminal-Bench 2.0 Leaderboard Trajectories dataset contains agent trajectories extracted from Terminal-Bench 2.0 leaderboard submissions. Each row includes a prompt (task instruction), the agents response, and the reward (pass/fail). The dataset features multiple configurations such as all trajectories, passed-only trajectories, ML/training-related tasks, and more. It also includes performance data for various models like Claude-Opus-4.6, Gemini-3.1-Pro-Preview, among others, along with data on ML tasks.