SmartBench (Context-Dependent) - Anomaly Localization: leaderboard
Metric: Anomaly location score (0-1 times 100): Jaccard similarity between the devices the model blames and the annotated evidence devices, averaged over the anomalous samples of the 2,400 context-dependent samples (1,200 normal, 1,200 anomalous compressed device-event sequences); higher is better. Source: arxiv.org. Saturation forecast: Around April 2028. 10 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 2.5 Pro | 36.5 | #145 |
| 2 | Gemini 3 Pro (Preview) | 34.7 | #64 |
| 3 | DeepSeek R1 | 26.1 | #245 |
| 4 | Claude Sonnet 4.5 | 25.7 | #138 |
| 5 | GPT-5 Mini | 25.2 | #176 |
| 6 | GPT-5 | 25.1 | #91 |
| 7 | Claude Sonnet 4 (20250514) | 24.7 | #211 |
| 8 | Qwen 3 32B | 18.5 | #424 |
| 9 | DeepSeek V3 | 17 | #312 |
| 10 | Qwen 3 8B | 10.5 | #667 |
No result here: #3 Claude Opus 5.5, #5 GPT-6 Astra, #8 Claude Fable 5.1.
Interactive version: theaggregate.ai/benchmark?slug=smartbench-context-dependent-anomaly-localization · How It Works · Data refreshed daily, snapshot 2026-10-11.