smirki/qwen3.5-4b-terminal-v1
The smirki/qwen3.5-4b-terminal-v1 is a 4.5 billion parameter variant of the Qwen3.5-4B model, developed by smirki, specifically optimized for terminal and coding-agent use. It maintains the same architecture, tokenizer, and chat template as the base Qwen model, including native XML tool-call format. This model demonstrates improved performance on the Terminal-Bench 2.1 benchmark, making it suitable for tasks requiring shell command execution and agent-based interactions.
Loading preview...
Overview
smirki/qwen3.5-4b-terminal-v1 is a specialized 4.5 billion parameter model derived from Qwen/Qwen3.5-4B, designed for enhanced performance in terminal environments and coding-agent applications. It utilizes the same Qwen3.5 architecture, tokenizer, and chat template, including its native XML tool-call format, making it a direct replacement for the base model in such use cases.
Key Capabilities
- Optimized for Terminal & Coding Agents: Specifically fine-tuned to excel in tasks involving shell command execution and agentic workflows.
- Improved Terminal-Bench Performance: Achieves a mean reward of 0.211 on Terminal-Bench 2.1, outperforming the base Qwen3.5-4B model (0.145). It also solved 21 out of 76 tasks at least once, compared to 15 for the base model.
- XML Tool-Call Support: Inherits the native XML tool-call format, compatible with vLLM's
qwen3_xmlparser for seamless integration. - Efficient Resource Usage: The
--language-model-onlyoption allows dropping the unused vision tower, optimizing resource allocation.
When to Use This Model
- Coding Agents: Ideal for applications where an AI agent needs to interact with a terminal, execute commands, and perform coding-related tasks.
- Automated Scripting: Suitable for automating system administration or development workflows through shell commands.
- Tool Use Scenarios: When leveraging the Qwen3.5 native XML tool-call format for structured function calling.