inclusionAI/Qwen3-32B-AWorld

Qwen3-32B-AWorld: Agentic LLM for Complex Tool Use

Qwen3-32B-AWorld, developed by inclusionAI, is a 32 billion parameter large language model fine-tuned from the Qwen3-32B base model. Its core specialization lies in advanced agent capabilities and proficient tool usage, designed to tackle complex, multi-step tasks requiring external interactions.

Key Capabilities

• Exceptional Agentic Performance: The model demonstrates strong performance in agent-based tasks, achieving a pass@1 score on the GAIA benchmark that exceeds GPT-4o and is comparable to DeepSeek-V3.
• Proficient Tool Integration: It is specifically optimized for precise integration and utilization of external tools, enabling it to perform actions like web searches and other function calls.
• Scalable Deployment: Recommended for deployment on 8 GPUs using vLLM, supporting efficient inference with tensor-parallel-size configuration.
• Dedicated Tool-Calling Flags: Requires specific vLLM flags (--enable-auto-tool-choice and --tool-call-parser hermes) to activate its tool-calling functionalities, ensuring correct processing and parsing of tool outputs.

Good For

• Automated Task Execution: Ideal for applications requiring autonomous agents to perform tasks by interacting with external systems or APIs.
• Complex Problem Solving: Suitable for scenarios where problems necessitate breaking down into sub-tasks and utilizing various tools to find solutions.
• Research and Development in Agent AI: Provides a robust foundation for exploring and building advanced AI agents.