alibaba-pai/SearchQwen3-8B
SearchQwen3-8B is an 8.19 billion parameter compact Search Agent model developed by Alibaba Cloud PAI, built upon the Qwen3-8B base model. It is specifically trained with environment-aligned, solver-verified search trajectories using EasyDistill 2.0, enabling robust multi-hop search, browsing, and evidence integration capabilities. This model excels at structured tool-call interactions for search tasks, demonstrating significantly improved accuracy in Deep Search and overall performance compared to its base model. With a context length of 40,960 tokens, it is optimized for complex information retrieval and agentic workflows.
Loading preview...
Overview
SearchQwen3-8B is an 8.19 billion parameter Search Agent model developed by Alibaba Cloud PAI, based on the Qwen3-8B architecture. It is specifically designed for agentic workflows involving search, browsing, and evidence integration, leveraging structured tool-call interfaces.
Key Capabilities
- Optimized for Search Agent Tasks: Trained using EasyDistill 2.0 with environment-aligned, solver-verified search trajectories from the SynSearch-Data dataset.
- Structured Tool-Call Support: Designed to interact effectively with external search and browse tools through a structured tool-call mechanism.
- Enhanced Performance: Demonstrates improved accuracy in Deep Search (35.42%) and overall performance (50.31%) on LLM-judge benchmarks compared to the base Qwen3-8B model, particularly in multi-hop QA and deep search tasks.
- Large Context Window: Features a context length of 40,960 tokens, suitable for processing extensive search results and complex queries.
When to Use This Model
SearchQwen3-8B is ideal for applications requiring a compact yet powerful model for:
- Multi-hop Question Answering: Effectively integrates information from multiple search queries to answer complex questions.
- Automated Browsing: Navigating and extracting information from web pages.
- Evidence Integration: Synthesizing information from various sources to form coherent responses.
- Agentic Workflows: Serving as the core intelligence for search-driven AI agents that need to interact with external tools.