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

#ModelScoreOverall rank
1GPT-544.6#178
2Gemini 3 Pro44.6#145
3Qwen 3 VL 32B (post-trained)41.1
4Claude Opus 4.637.5#113
5GPT-4o35.7#751
6GPT-4.133.9#505
7Qwen 3 VL 32B33.9#622
8Claude Sonnet 4.530.4#261
9Toto-1.0-QA-Experimental 32B (TSFM-VLM)26.8
10Qwen 3 VL 8B25#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.