WebExplorer: Explore and Evolve for Training Long-Horizon Web Agents

WebExplorer is a collaboration between HKUST and MiniMax.

Abstract

WebExplorer introduces a systematic approach for training long-horizon web agents through model-based exploration and iterative query evolution. Our method generates challenging query-answer pairs requiring multi-step reasoning and complex web navigation, achieving state-of-the-art performance at 8B parameter scale.

🔥 News

• [2025/09/22] :fire: We released the inference code and evaluation framework.

• [2025/09/08] :fire: We released our model WebExplorer-8B and dataset WebExplorer-QA with 100 high-quality query-answer pairs!

📊 Key Results

Model	BC-en	BC-zh	GAIA	WebWalkerQA	FRAMES	Xbench-DS	HLE
OpenAI-o3†	50.9	58.1	70.5†	71.7	84.0	66.7	20.2
Claude-4-Sonnet†	12.2	29.1	68.3†	61.7	80.7	64.6	20.3
GLM-4.5	26.4	37.5	66.0†	65.6†	78.9†	70.0†	21.2†
DeepSeek-V3.1	30.0	49.2	63.1†	61.2†	83.7	71.2	29.8
Kimi-K2†	14.1	28.8	57.7	63.0	72.0	50.0	18.1
====	====	====	====	====	====	====	====
WebShaper-72B	-	-	60.0	52.2	-	-	-
WebShaper-32B (QwQ)	-	-	53.3	49.7	-	-	-
WebShaper-32B	-	-	52.4	51.4	-	-	-
WebSailor-72B	12.0	30.1	55.4	-	-	55.0	-
WebSailor-32B	10.5	25.5	53.2	-	-	53.3	-
WebSailor-7B	6.7	14.2	33.0	-	-	34.3	-
ASearcher-Web-QwQ	5.2	15.6	52.8	34.3	70.9	42.1	12.5
AFM-32B	11.1	-	55.3	63.0	-	-	18.0
AFM-7B	5.8	-	40.8	55.6	-	-	15.6
WebThinker-32B	2.8	-	48.5	46.5	-	-	15.8
MiroThinker-32B-DPO-v0.1	13.0	17.0	57.3	49.3	71.7	-	11.8
MiroThinker-8B-DPO-v0.1	8.7	13.6	46.6	45.7	64.4	-	-
WebExplorer-8B (SFT)	7.9	21.3	43.7	59.8	72.6	47.5	16.0
WebExplorer-8B (RL)	15.7	32.0	50.0	62.7	75.7	53.7	17.3

Accuracy (%) of web agents on information-seeking benchmarks. BC-en and BC-zh denote BrowseComp-en and BrowseComp-zh respectively. XBench-DS refers to XBench-DeepSearch. Bold indicates the best performance among open-source models < 100B, while underlined values represent the best performance among models < 10B parameters. All scores of WebExplorer-8B are computed as Avg@4 using LLM-as-Judge. Entries marked with a dagger (†) were reproduced by us under our scaffold: on model name = entire row; on a number = that entry only.

✨ Key Features

• 🌐 Long-horizon Reasoning: Supports up to 128K context length and 100 tool calling turns
• 🛠️ Tool Utilization: Masters search and browse functionalities
• 🏆 State-of-the-art Performance: Achieves best-in-class results among models under 10B parameters

🚀 Resources

🤗 Models

Model Name	Size	Description	Link
WebExplorer-8B	8B	Long-horizon web agent trained on WebExplorer-QA	🤗 HuggingFace

📚 Datasets

Dataset Name	Size	Description	Link
WebExplorer-QA	100 samples	High-quality query-answer pairs for web agent training	🤗 HuggingFace

🛠️ Tool Schema

WebExplorer-8B supports two tools for web interaction:

1. Browse Tool

{
    "name": "browse",
    "type": "function",
    "description": "Extract specific information from a webpage",
    "parameters": {
        "type": "object",
        "properties": {
            "url": {
                "type": "string",
                "description": "Target URL to browse. The webpage content will be processed by the LLM for information extraction."
            },
            "query": {
                "type": "string",
                "description": "Specific query about the webpage content. The LLM will analyze the content to answer this query."
            }
        },
        "required": ["url", "query"]
    }
}

2. Search Tool

{
    "name": "search",
    "type": "function",
    "description": "Perform web search queries",
    "parameters": {
        "type": "object",
        "properties": {
            "queries": {
                "type": "array",
                "items": {
                    "type": "string"
                },
                "description": "List of search queries. Returns search results containing title, URL, and snippet for each query."
            }
        },
        "required": ["queries"]
    }
}

📝 Citation

If you find our work useful, please consider citing:

@misc{liu2025webexplorer,
      title={WebExplorer: Explore and Evolve for Training Long-Horizon Web Agents}, 
      author={Junteng Liu and Yunji Li and Chi Zhang and Jingyang Li and Aili Chen and Ke Ji and Weiyu Cheng and Zijia Wu and Chengyu Du and Qidi Xu and Jiayuan Song and Zhengmao Zhu and Wenhu Chen and Pengyu Zhao and Junxian He},
      year={2025},
      eprint={2509.06501},
      archivePrefix={arXiv},
      primaryClass={cs.CL},
      url={https://arxiv.org/abs/2509.06501}, 
}