reasonwang/SkillGym-Qwen3.5-4B
SkillGym-Qwen3.5-4B is a 4.5 billion parameter language model developed by Reason Wang, fine-tuned from Qwen3.5-4B. It is specifically trained on SkillGym trajectories to enable agents to effectively use skills, which are folders of instructions, documents, and scripts. This model excels at applying relevant skills through tool calls or shell commands within a sandboxed environment, making it suitable for agentic task execution rather than general chat assistance. It features a 32768-token context length and is optimized for verifiable environment generation.
Loading preview...
SkillGym-Qwen3.5-4B Overview
SkillGym-Qwen3.5-4B is a 4.5 billion parameter model, fine-tuned from the Qwen3.5-4B base model by Reason Wang. Its primary purpose is to enable AI agents to effectively utilize "skills"—structured folders containing instructions, reference documents, and scripts—to solve tasks. This model was developed as part of the research presented in the paper "SkillGym: Training Skill-Use Agents with Automatic Verifiable Environment Generation" and is specifically designed for agentic workflows.
Key Capabilities
- Skill-Based Agentic Execution: Trained to interpret and apply agent skills, executing tasks through tool calls or shell commands in a sandboxed environment.
- Enhanced Performance on Agent Benchmarks: Demonstrates significant improvements over its base model on various agent-specific benchmarks, including SkillGym (47.0 vs 33.8), SkillEval (70.8 ± 1.6 vs 62.8 ± 1.5), SkillsBench (14.3 ± 1.0 vs 10.1 ± 1.3), and Skill-Use-Bench (48.7 vs 8.8).
- Reasoning Traces: Incorporates reasoning traces during its supervised fine-tuning process, which involved 19k verified successful trajectories from three teacher models.
Good For
- Developing Skill-Enabled Agents: Ideal for researchers and developers building agents that need to interact with environments using predefined skills and tools.
- Automated Task Execution: Suitable for scenarios requiring agents to perform complex tasks by consulting and applying specific instructions and scripts.
- Research in Agentic AI: A valuable model for exploring and advancing the field of agent skill utilization and verifiable environment generation.