Terminal Bench v4 Scores Released Comparing Model Command-Line Performance

A community post shared the latest Terminal Bench v4 scores evaluating the terminal command execution capabilities of various open and closed LLMs. GLM-5.3 secured the top position with a score of 41.9%, followed by GLM-5.3-Flash at 32.8% and DSV4.1-Flash at 26.8%. Terminal execution benchmarks are increasingly viewed as reliable indicators of practical AI agent capabilities and real-world task execution. The results highlight how open-weights models like the GLM series are making rapid strides in practical coding and system operations. While GLM-5.3 dominated the evaluation, smaller models generally struggled, with Qwen3.8-27B scoring just 5.6% and gemma4-31b failing to score any points. Furthermore, Kimi-K3 underperformed relative to its model size, achieving only a 12.6% score.