WebWalkerQA
Qwen2.5-72B 33.3% leads the models we track Agents start from a website's home page and traverse its subpages to find the facts that answer a question. Answers are judged against references.
Tongyi deep research model reports cite it. 23 upvotes on Hugging Face papers.
- Made by
- Alibaba Tongyi Lab
- Released
- January 2025
- Kind
- Deep research
- Level
- Everyday
- Scored
- Graded by a model
- Answers
- Short answer
- Tasks
- All public
- Strongest model tested
- o1-preview #147 on the capability index
- Last model added
- September 2024
- Scores refreshed
- November 12th, 2026
In the authors’ words
Retrieval-augmented generation (RAG) demonstrates remarkable performance across tasks in open-domain question-answering. However, traditional search engines may retrieve shallow content, limiting the ability of LLMs to handle complex, multi-layered information. To address it, we introduce WebWalkerQA, a benchmark designed to assess the ability of LLMs to perform web traversal. It evaluates the capacity of LLMs to traverse a website's subpages to extract high-quality data systematically. We propose WebWalker, which is a multi-agent framework that mimics human-like web navigation through an explore-critic paradigm. Extensive experimental results show that WebWalkerQA is challenging and demonstrates the effectiveness of RAG combined with WebWalker, through the horizontal and vertical integration in real-world scenarios.
Tongyi (paper system, backbone unspecified) tops the leaderboard with 40.7%
Best score per entry from the benchmark's own leaderboard, read on November 12th, 2026. Where a model ran with several settings or harnesses, we keep its best. Names in gray aren't in our model list: fine-tunes, agent systems, products and older models. They don't count toward leaders. The best model we track is Qwen2.5-72B, at 33.3%.
The newest model with a score here came out in September 2024, and nobody has tested a newer one. Among the 20 highest-ranked models on the capability index, these have no score yet:
- Claude Fable 5.2
-
Gemini 4 Carbon - GPT-6 Astra
- Claude Opus 5.5
- and 16 more
- 40.7 Tongyi (paper system, backbone unspecified)
- 37.5 GPT-4o (snapshot unspecified)
- 37.4 Kimi (paper system, backbone unspecified)
- 33.8 Qwen-Plus (snapshot unspecified)
- 33.3 Qwen2.5-72B
- 29.0 ERNIE-4.0-8K
- 27.9 Gemini-Search (paper system, backbone unspecified)
- 27.5 Qwen2.5-14B
- 26.0 Qwen2.5-32B
- 20.7 Naive RAG (paper system, backbone unspecified)
- 19.9 Qwen2.5-7B
- 16.8 Doubao (paper system, backbone unspecified)
- 11.3 MindSearch (paper system, backbone unspecified)
- 9.85 o1-preview
- 8.08 Gemini-1.5-Pro (snapshot unspecified)
- 1 Tongyi (paper system, backbone unspecified) retrieval system 40.7%
- 2 GPT-4o (snapshot unspecified) 37.5%
- 3 Kimi (paper system, backbone unspecified) retrieval system 37.4%
- 4 Qwen-Plus (snapshot unspecified) 33.8%
- 5 Qwen2.5-72B 33.3%
- 6 ERNIE-4.0-8K retrieval system 29%
- 7 Gemini-Search (paper system, backbone unspecified) retrieval system 27.9%
- 8 Qwen2.5-14B 27.5%
- 9 Qwen2.5-32B 26%
- 10 Naive RAG (paper system, backbone unspecified) retrieval system 20.7%
- 11 Qwen2.5-7B 19.9%
- 12 Doubao (paper system, backbone unspecified) retrieval system 16.8%
- 13 MindSearch (paper system, backbone unspecified) retrieval system 11.3%
- 14 o1-preview closed book 9.85%
- 15 Gemini-1.5-Pro (snapshot unspecified) closed book 8.08%
Source: Official WebWalkerQA leaderboard
| # | Entry | Score | Setting | Date | Source |
|---|---|---|---|---|---|
| 1 | Tongyi (paper system, backbone unspecified) | 40.7% | retrieval system | Official leaderboard | |
| 2 | GPT-4o (snapshot unspecified) | 37.5% | WebWalker | Official leaderboard | |
| 3 | Kimi (paper system, backbone unspecified) | 37.4% | retrieval system | Official leaderboard | |
| 4 | Qwen-Plus (snapshot unspecified) | 33.8% | WebWalker | Official leaderboard | |
| 5 | Qwen2.5-72B | 33.3% | WebWalker | Sep 2024 | Official leaderboard |
| 6 | ERNIE-4.0-8K | 29% | retrieval system | Official leaderboard | |
| 7 | Gemini-Search (paper system, backbone unspecified) | 27.9% | retrieval system | Official leaderboard | |
| 8 | Qwen2.5-14B | 27.5% | WebWalker | Sep 2024 | Official leaderboard |
| 9 | Qwen2.5-32B | 26% | WebWalker | Sep 2024 | Official leaderboard |
| 10 | Naive RAG (paper system, backbone unspecified) | 20.7% | retrieval system | Official leaderboard | |
| 11 | Qwen2.5-7B | 19.9% | WebWalker | Sep 2024 | Official leaderboard |
| 12 | Doubao (paper system, backbone unspecified) | 16.8% | retrieval system | Official leaderboard | |
| 13 | MindSearch (paper system, backbone unspecified) | 11.3% | retrieval system | Official leaderboard | |
| 14 | o1-preview | 9.85% | closed book | Sep 2024 | Official leaderboard |
| 15 | Gemini-1.5-Pro (snapshot unspecified) | 8.08% | closed book | Official leaderboard |
Number one over time
Lab claims count from the day the card came out, independent runs from the model’s release.
Source: Official WebWalkerQA leaderboard
Scores over time
All the scores we have, one dot each, so a model tested at several settings appears more than once.
Source: Official WebWalkerQA leaderboard
No lab model card reports it yet
We read the headline results of 522 launch posts from 52 labs, and none of them list this benchmark.
More in Tool use
- AutomationBench In use Agent completes end-to-end business workflows inside a simulated company exposing dozens of REST APIs. Graded on resulting state. Gemini 4 Carbon 53.9%
- MCP Atlas In use Multi-step workflows over real MCP servers. Scores tool-use competency on realistic tasks. Gemini 4 Argon 90.7%
- BFCL V4 In use Models call functions and tools, including multi-turn agentic tasks with web search and memory. Checked against expected calls and outcomes. Qwen3.7-Max 75%
- τ³-bench Banking In use Agents act as banking customer-service reps over many turns with a simulated user, following policy. Pass^1 on the final database state. Qwen 3.8 Max 55.1%
- Toolathlon-Verified In use 108 long-horizon tasks across 32 apps through about 600 MCP tools. Execution-based checkers inspect final environment state. GLM-5.3-Flash 78.4%
- DeepSearchQA In use Questions that need multi-step search and synthesis. Final answers graded against references. Kimi K3 95%