Nemotron-Terminal-14B Overview

Nemotron-Terminal-14B is a 14 billion parameter model from the Nemotron-Terminal family, developed by NVIDIA. It is fine-tuned from the Qwen3 architecture and specifically designed for autonomous terminal interaction. The model was trained using the extensive Nemotron-Terminal-Corpus dataset, which focuses on terminal-specific tasks.

Key Capabilities

• Specialized for Terminal Tasks: Demonstrates significant performance improvements over its Qwen3 baseline on the Terminal-Bench 2.0, achieving 20.2% accuracy compared to the baseline's 4.04%.
• Structured Output: Generates structured JSON outputs, including analysis, step-by-step plans, and specific keystroke commands, facilitating automated execution.
• Terminus 2 Scaffolding: Utilizes the Terminus 2 scaffolding for training and is recommended for consistent evaluation.

Good for

• Automating complex command-line operations.
• Developing agents for system administration and task execution in terminal environments.
• Research and development in autonomous computing and human-computer interaction via command-line interfaces.