TSQueryBench - Independent Scoring: leaderboard

Grading single time-series explanations on TSQueryBench (2026): the model labels each of 1,500 rule-built explanations of synthetic series as correct, partially correct or incorrect without seeing the alternatives. Accuracy against the construction labels in percent; higher is better.

Metric: Independent scoring accuracy (%, times 100): share of the 1,500 candidate explanations whose correctness label (correct, partially correct, incorrect) the model assigns correctly when it sees one explanation at a time on TSQueryBench, 500 synthetic time series in ten query types (spikes, drops, structural breaks, mean and volatility shifts, trends, ordering, shape), zero-shot with structured JSON output; higher is better. Source: arxiv.org. Saturation forecast: Around February 2027. 6 models tracked. Updated 7 Oct 2026.

Top models

#ModelScoreOverall rank
1Qwen 3 8B72#1873
2Gemma 4 31B (IT)67#350
3DeepSeek V3.256#419
4Gemma 3 12B (IT)47#1887
5Claude 3 Haiku35#2663
6Llama 3.1 8B Instruct29#4600

No result here: #1 Claude Fable 5.2, #3 Gemini 4 Neon, #4 GPT-6.2 Astra.

Interactive version: theaggregate.ai/benchmark?slug=tsquerybench-independent-scoring · How It Works · Data refreshed daily, snapshot 2026-11-21.