TerminalBench 2.1 benchmark features GPT-5.6 variants

TerminalBench 2.1 evaluates AI agents in terminal environments using GPT-5.6 Sol, Terra, and Luna models. The benchmark focuses on task completion and efficiency across diverse command-line scenarios.