TSQueryBench - Explanation Generation: leaderboard
Numerically grounded explanation of time series on TSQueryBench (2026): the model writes an explanation for one of 500 synthetic series in ten query types (spikes, structural breaks, mean and volatility shifts, trends and others), and an annotator checks every numerical claim against the data. Percent of explanations judged fully correct; higher is better.
Metric: Explanation generation accuracy (%, times 100): share of the model's free-form explanations rated fully correct by a domain-knowledgeable annotator who verifies every numerical claim against the series 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 2029. 6 models tracked. Updated 7 Oct 2026.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Qwen 3 8B | 38 | #1873 |
| 2 | Gemma 4 31B (IT) | 36 | #350 |
| 3 | Claude 3 Haiku | 31 | #2663 |
| 4 | DeepSeek V3.2 | 30 | #419 |
| 5 | Llama 3.1 8B Instruct | 18 | #4600 |
| 6 | Gemma 3 12B (IT) | 13 | #1887 |
No result here: #1 Claude Fable 5.2, #3 Gemini 4 Neon, #4 GPT-6.2 Astra.
Interactive version: theaggregate.ai/benchmark?slug=tsquerybench-explanation-generation · How It Works · Data refreshed daily, snapshot 2026-11-21.