TSQueryBench - Anomaly Count Accuracy: leaderboard

Counting anomalies in synthetic time series on TSQueryBench (2026): 100 series of length 100 to 500 hold between one and ten injected anomalies, and the model must report them. Percent of series where the reported count is exactly right; higher is better.

Metric: Multi-anomaly count accuracy (%, times 100): share of 100 synthetic series (1 to 10 anomalies, lengths 100 to 500) where the model reports exactly the right number of anomalies, zero-shot; higher is better. Source: arxiv.org. Saturation forecast: Around February 2027. 6 models tracked. Updated 7 Oct 2026.

Top models

#ModelScoreOverall rank
1Gemma 4 31B (IT)45#354
2DeepSeek V3.225#425
3Qwen 3 8B12#1890
4Llama 3.1 8B Instruct1#4655
5Claude 3 Haiku1#2683
6Gemma 3 12B (IT)0#1907

No result here: #2 Claude Fable 5.2, #3 Gemini 4 Neon, #4 GPT-6.2 Astra.

Interactive version: theaggregate.ai/benchmark?slug=tsquerybench-anomaly-count-accuracy · How It Works · Data refreshed daily, snapshot 2026-11-22.