AgentWebBench (LLM-Selected Tools) - Question Answering: leaderboard
AgentWebBench (2026) models the Agentic Web: a user agent serves information needs over 100 ClueWeb22-B websites by coordinating with content agents or tools that each cover one website, with every LLM in thinking mode. This board is answer accuracy in percent judged by an LLM on DeepResearchGym short-answer questions, under LLM-selected tools, the user agent choosing websites from their descriptions and querying a per-site retrieval tool; higher is better.
Metric: Answer accuracy (%, LLM judge) on 53 DeepResearchGym short-answer questions covered by the corpus, LLM user agent under tool-based prompted selection (the user agent chooses websites from their descriptions and calls a per-website dense retrieval tool, no content agents), thinking mode on; higher is better. Source: arxiv.org. Saturation forecast: Around February 2027. 7 models tracked. Updated 7 Oct 2026.
Top models
| # | Model | Setting | Score | Overall rank |
|---|---|---|---|---|
| 1 | Gemini 3 Flash (Preview) | 67.92 | #152 | |
| 2 | GPT-5 Mini | 32.08 | #364 | |
| 3 | Qwen 3 Next 80B A3B (Thinking) | 32.08 | #594 | |
| 4 | DeepSeek V3.2 | Thinking | 24.53 | #419 |
| 5 | Qwen 3 30B A3B 2507 (Thinking) | 24.53 | #764 | |
| 6 | Qwen 3 14B | 20.75 | #1269 | |
| 7 | Qwen 3 4B 2507 (Thinking) | 13.21 | #1433 |
No result here: #1 Claude Fable 5.2, #3 Gemini 4 Neon, #4 GPT-6.2 Astra.
Interactive version: theaggregate.ai/benchmark?slug=agentwebbench-llm-selected-tools-question-answering · How It Works · Data refreshed daily, snapshot 2026-11-21.