AgentNoiseBench (Search, Tool Noise): leaderboard
Metric: Accuracy on 500 sampled 2WikiMultiHopQA and HotpotQA questions answered with a search tool (mean of the two) with tool noise injected into tool returns (execution failures, incomplete, erroneous, misleading or redundant outputs); noise produced by a fixed generator whose prompt was optimized to degrade a reference agent while keeping every task solvable; task success rate (fraction times 100), mean of four trials; the paper's stability-gated criterion counts a task solved only with a correct final outcome and no noise-induced deviation in the trajectory; higher is better. Source: arxiv.org. Saturation forecast: Around 2029. 24 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GLM-4.6 (Non-reasoning) | 39 | #246 (GLM-4.6) |
| 2 | GPT-5.2 (Thinking) | 37 | #105 (GPT-5.2) |
| 3 | DeepSeek V3.1 (Non-reasoning) | 37 | #260 (DeepSeek V3.1) |
| 4 | GLM-4.5 (Thinking) | 37 | #265 (GLM-4.5) |
| 5 | DeepSeek V3.2 Exp (Non-reasoning) | 36 | #227 (DeepSeek V3.2 Exp) |
| 6 | GLM-4.6 (Thinking) | 36 | #246 (GLM-4.6) |
| 7 | GLM-4.5 (Non-reasoning) | 36 | #265 (GLM-4.5) |
| 8 | O3 | 35 | #121 |
| 9 | Qwen 3 Max (Thinking) | 33 | #201 (Qwen 3 Max) |
| 10 | Claude Sonnet 4 (Thinking) | 31 | #194 (Claude Sonnet 4) |
| 11 | Gemini 2.5 Pro | 30 | #145 |
| 12 | Claude Sonnet 4 | 30 | #194 |
| 13 | LongCat-Flash-Chat | 30 | #322 |
| 14 | Claude Sonnet 4.5 | 29 | #138 |
| 15 | Claude Sonnet 4.5 (Thinking) | 29 | #138 (Claude Sonnet 4.5) |
No result here: #3 Claude Opus 5.5, #5 GPT-6 Astra, #8 Claude Fable 5.1.
Interactive version: theaggregate.ai/benchmark?slug=agentnoisebench-search-tool-noise · How It Works · Data refreshed daily, snapshot 2026-10-11.