Qwen3-30B-A3B-Thinking-2507-Researcher

Qwen3-30B-A3B-Thinking-2507-Researcher is a research-focused adaptation of Qwen3-30B-A3B-Thinking-2507 developed in the DeepResearch framework. It is intended for multi-step information gathering, evidence synthesis, and tool-augmented reasoning workflows.

Overview

This model is designed for:

• research assistance
• literature exploration
• evidence gathering
• structured reasoning with tools

Training

This model was trained using GRPO on the following datasets:

• novonordisk-red/pmc-qa-hard
• novonordisk-red/springernature-medium-qa

Benchmarks

Internal benchmark results can be summarized here.

Limitations

• May produce incorrect or incomplete claims
• Depends on prompt quality, retrieval quality, and tool setup
• Should be validated before use in high-stakes settings

Base Model

• Qwen/Qwen3-30B-A3B-Thinking-2507