ARFBench - Start Time: leaderboard
Time series question answering on software incident data in ARFBench (2026, Datadog): multiple-choice questions over multivariate production telemetry from real incidents, with the series shown as chart images. Accuracy on the Start Time questions (when the first anomaly begins), in percent; higher is better.
Metric: Accuracy (%) on the 56 Start Time questions (when the first anomaly begins, or no anomaly) of ARFBench, for the time series rendered as chart images (Matplotlib plots, three images for paired questions), with the series description and channel names as text; few-shot multiple-choice prompting with shuffled answer choices, temperature 0.05 (default for models without the setting), reasoning models at medium effort with 2,000 output tokens, blank or invalid answers counted wrong; higher is better. Source: arxiv.org. Saturation forecast: Around April 2028. 10 models tracked. Updated 7 Oct 2026.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5 | 44.6 | #178 |
| 2 | Gemini 3 Pro | 44.6 | #145 |
| 3 | Qwen 3 VL 32B (post-trained) | 41.1 | |
| 4 | Claude Opus 4.6 | 37.5 | #113 |
| 5 | GPT-4o | 35.7 | #751 |
| 6 | GPT-4.1 | 33.9 | #505 |
| 7 | Qwen 3 VL 32B | 33.9 | #622 |
| 8 | Claude Sonnet 4.5 | 30.4 | #261 |
| 9 | Toto-1.0-QA-Experimental 32B (TSFM-VLM) | 26.8 | |
| 10 | Qwen 3 VL 8B | 25 | #917 |
No result here: #1 Claude Fable 5.2, #3 Gemini 4 Neon, #4 GPT-6.2 Astra.
Interactive version: theaggregate.ai/benchmark?slug=arfbench-start-time · How It Works · Data refreshed daily, snapshot 2026-11-21.