TS-Skill (MCQ): leaderboard
Time-series question answering benchmark (TS-Skill, Han et al., UCLA and Carnegie Mellon University, 2026) of 3,000 questions about synthetic, timestamped series from 25 domains, each labelled with the skills it needs: temporal scale selection, temporal localization and cross-interval integration. This is the multiple-choice protocol: four options (two for yes/no questions), a reply of a single letter, and 0 for an unparseable reply; the score is macro-F1 on a 0-1 scale, higher is better. Text models read the series as (timestamp, value) pairs and vision-language models also get a rendered plot.
Metric: MCQ macro-F1 (0-1; four-option multiple choice, letter-only answers, unparseable answers score 0, 3,000 questions). Source: arxiv.org. Saturation forecast: Estimated already saturated. 9 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Sonnet 4.5 | 0.78 | #244 |
| 2 | GPT-4o | 0.77 | #741 |
| 3 | Qwen 3 32B (Thinking) | 0.69 | #960 (Qwen 3 32B) |
| 4 | Qwen 2.5 14B Instruct | 0.58 | #1745 |
| 5 | Qwen 2.5 VL 7B | 0.56 | #1796 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=ts-skill-mcq · How It Works · Data refreshed daily, snapshot 2026-11-12.