Research map / Time series / Foundation models for time series
TSRBench: A Comprehensive Multi-task Multi-modal Time Series Reasoning Benchmark for Generalist Models
Fangxu Yu, Xingang Guo, Lingzhi Yuan, Haoqiang Kang, Hongyu Zhao, Lianhui Qin, Furong Huang, Bin Hu, Tianyi Zhou
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| scale up model size | open-source LLMs · TSRBench overall | Spearman rho with size: 0.9248 | Improved |
| scale up model size | open-source LLMs and VLMs · TSRBench Prediction tasks | Spearman rho with size: -0.2415 (LLM), -0.2612 (VLM) | No clear effect |
| use both text and image time series (T+V) instead of single modality (ablation) | GPT-5 · TSRBench overall | accuracy: 55.5 (T) / 52.4 (V) -> 55.6 (T+V) | No clear effect |
| use visual instead of textual time series (ablation) | GPT-5 · TSRBench Perception (Pattern Recognition) | accuracy: 75.7 -> 83.6 | Improved |
| enable high reasoning effort (ablation) | GPT-5-mini (T+V) · TSRBench overall | accuracy: 46.9 -> 54.1 | Improved |
| enable high reasoning effort (ablation) | o4-mini (T+V) · TSRBench overall | accuracy: 48.2 -> 52.5 | Improved |
| enable reasoning mode (ablation) | GPT-5 (T+V) · TSRBench overall | accuracy: 45.3 -> 55.6 | Improved |
| add time series analysis tool use (ablation) | o4-mini (T+V) · TSRBench overall | accuracy: +1.2 | Mixed |
| add time series analysis tool use (ablation) | GPT-5 (T+V) · TSRBench overall | accuracy: +0.6 | No clear effect |
| raise plot resolution from 10 to 100 PPI (ablation) | GPT-5 (V) · TSRBench overall | accuracy: 45.5 -> 52.4 | Improved |
| raise plot resolution from 100 to 400 PPI (ablation) | GPT-5 (V) · TSRBench overall | accuracy: 52.4 -> 52.3 | No clear effect |
Closest papers by meaning
- TimeXL: Explainable Multi-modal Time Series Prediction with LLM-in-the-Loop (NeurIPS 2025 · similarity 0.91)
- Rethinking Post-Training Recipes for Multimodal Time-Series Forecasting (NeurIPS 2026 · similarity 0.90)
- Time-MMD: Multi-Domain Multimodal Dataset for Time Series Analysis (NeurIPS 2024 · similarity 0.88)
- TS-RAG: Retrieval-Augmented Generation based Time Series Foundation Models are Stronger Zero-Shot Forecaster (NeurIPS 2025 · similarity 0.88)
- Time-LLM: Time Series Forecasting by Reprogramming Large Language Models (ICLR 2024 · similarity 0.87)
- TimeBase: The Power of Minimalism in Efficient Long-term Time Series Forecasting (ICML 2025 · similarity 0.87)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
