AgentNoiseBench (VitaBench, User Noise): leaderboard
Metric: Success on a stratified 25 percent sample of VitaBench tasks (delivery and in-store) with user noise injected into the simulated user's turns (ambiguous, inconsistent, redundant, topic-drifting or boundary-probing instructions); 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 2028. 24 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5.2 (Thinking) | 45 | #105 (GPT-5.2) |
| 2 | O3 | 44 | #121 |
| 3 | Claude Sonnet 4.5 (Thinking) | 41 | #138 (Claude Sonnet 4.5) |
| 4 | Qwen 3 Max (Thinking) | 40 | #201 (Qwen 3 Max) |
| 5 | GLM-4.6 (Thinking) | 40 | #246 (GLM-4.6) |
| 6 | Claude Sonnet 4.5 | 39 | #138 |
| 7 | Gemini 2.5 Pro | 38 | #145 |
| 8 | GLM-4.5 (Thinking) | 37 | #265 (GLM-4.5) |
| 9 | DeepSeek R1 0528 | 36 | #217 |
| 10 | Claude Sonnet 4 (Thinking) | 36 | #194 (Claude Sonnet 4) |
| 11 | LongCat-Flash-Chat | 36 | #322 |
| 12 | GPT-4.1 | 33 | #240 |
| 13 | Claude Sonnet 4 | 33 | #194 |
| 14 | GLM-4.6 (Non-reasoning) | 33 | #246 (GLM-4.6) |
| 15 | DeepSeek V3.2 Exp (Non-reasoning) | 32 | #227 (DeepSeek V3.2 Exp) |
No result here: #3 Claude Opus 5.5, #5 GPT-6 Astra, #8 Claude Fable 5.1.
Interactive version: theaggregate.ai/benchmark?slug=agentnoisebench-vitabench-user-noise · How It Works · Data refreshed daily, snapshot 2026-10-11.