mlsanityData to 14 Oct 2026

TSRouter: Dynamic Modality-Model Selection for Time Series Reasoning

arXiv 2607.08940 ↗First posted 9 Jul 2026cs.LGcited by 2

On the map: LLM reasoning → Time series reasoning

Abstract (arXiv)

Time series reasoning is essential for real-world problem-solving. While both Large Language Models (LLMs) and Vision-Language Models (VLMs) can reason about time-series data, their capabilities are complementary: LLMs process time series as text sequences and thus preserve exact numerical understanding, but struggle with global patterns, whereas VLMs efficiently capture these patterns by visualizing time series but may lose fine-grained details. Moreover, models vary significantly in task-specific expertise and inference costs. Dynamically selecting the most suitable modality and model for each query is therefore crucial, yet challenging because it requires modeling the complex interactions among tasks, queries, modalities, and models, which carry rich contextual signals. To this end, we introduce TSRouter, a graph-based dynamic routing framework. TSRouter constructs a heterogeneous graph of task, query, modality, and model nodes to contextualize the interactions among query characteristics, modality attributes, and model capabilities. TSRouter formulates routing as a candidate scoring problem, where each modality-model pair is evaluated based on user-defined performance-cost preferences to select the optimal candidate. Comprehensive evaluations on 4 distinct time series reasoning tasks reveal that TSRouter substantially outperforms diverse baselines with 16% to 46% relative improvements. Furthermore, TSRouter demonstrates robust zero-shot plug-and-play generalization to unseen models and novel tasks and preserves high performance while reducing computational overhead through cost-aware optimization. Our code is available at https://github.com/tianyi-lab/TSRouter.

Key facts. Cited by 2 papers in the map (data through 14 Oct 2026). Placed in Time series reasoning, one of 17 directions in LLM reasoning. Builds on 58 of its 69 references that are in the map. First posted on arXiv 9 Jul 2026.

Builds on

58 of 69 references.

Closest ancestors · 7

Same direction, newest first

  1. 2026TSRBench: A Comprehensive Multi-task Multi-modal Time Series Reasoning Benchmark for Generalist Models26 Jan
  2. 2025TS-Reasoner: Aligning Time Series Foundation Models with LLM Reasoning3 Oct
  3. 2025MTBench: A Multimodal Time Series Benchmark for Temporal Reasoning and Question Answering21 Mar
  4. 2025Achieving Time Series Reasoning Requires Rethinking Model Design, Tasks Formulation, and Evaluation3 Feb
  5. 2024ChatTS: Aligning Time Series with LLMs via Synthetic Data for Enhanced Understanding and Reasoning4 Dec
  6. 2024A Picture is Worth A Thousand Numbers: Enabling LLMs Reason about Time Series via Visualization9 Nov
  7. 2024Language Models Still Struggle to Zero-shot Reason about Time Series17 Apr

Other references · 51

  1. 2024The Llama 3 Herd of Modelscited by 14,715
  2. 2022Chain-of-Thought Prompting Elicits Reasoning in Large Language Modelscited by 12,459
  3. 2016Semi-Supervised Classification with Graph Convolutional Networkscited by 11,361
  4. 2025Qwen3 Technical Reportcited by 8,820
  5. 2025DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via Reinforcement Learningcited by 6,330
  6. 2023Tree of Thoughts: Deliberate Problem Solving with Large Language Modelscited by 3,211
  7. 2025Qwen3-VL Technical Reportcited by 2,793
  8. 2019Fast Graph Representation Learning with PyTorch Geometriccited by 2,434
  9. 2021Autoformer: Decomposition Transformers with Auto-Correlation for Long-Term Series Forecastingcited by 1,676
  10. 2022Least-to-Most Prompting Enables Complex Reasoning in Large Language Modelscited by 1,407
  11. 2025Qwen3 Embedding: Advancing Text Embedding and Reranking Through Foundation Modelscited by 1,261
  12. 2024Large Language Monkeys: Scaling Inference Compute with Repeated Samplingcited by 991
38 more
  1. 2023Graph of Thoughts: Solving Elaborate Problems with Large Language Modelscited by 886
  2. 2023Reasoning with Language Model is Planning with World Modelcited by 861
  3. 2019Temporal Fusion Transformers for Interpretable Multi-horizon Time Series Forecastingcited by 761
  4. 2023Time-LLM: Time Series Forecasting by Reprogramming Large Language Modelscited by 700
  5. 2024RouteLLM: Learning to Route LLMs with Preference Datacited by 682
  6. 2026Kimi K2.5: Visual Agentic Intelligencecited by 501
  7. 2025GLM-4.5V and GLM-4.1V-Thinking: Towards Versatile Multimodal Reasoning with Scalable Reinforcement Learningcited by 466
  8. 2026Self-Distilled Reasoner: On-Policy Self-Distillation for Large Language Modelscited by 455
  9. 2024Hybrid LLM: Cost-Efficient and Quality-Aware Query Routingcited by 372
  10. 2026Rethinking On-Policy Distillation of Large Language Models: Phenomenology, Mechanism, and Recipecited by 299
  11. 2026Self-Distillation Enables Continual Learningcited by 247
  12. 2024RouterBench: A Benchmark for Multi-LLM Routing Systemcited by 238
  13. 2023TEMPO: Prompt-based Generative Pre-trained Transformer for Time Series Forecastingcited by 174
  14. 2023Large Language Model Routing with Benchmark Datasetscited by 172
  15. 2024GraphRouter: A Graph-based Router for LLM Selectionscited by 165
  16. 2026Self-Distilled RLVRcited by 131
  17. 2023Let's Sample Step by Step: Adaptive-Consistency for Efficient Reasoning and Coding with LLMscited by 115
  18. 2024ChatTime: A Unified Multimodal Time Series Foundation Model Bridging Numerical and Textual Datacited by 100
  19. 2025Reasoning with Sampling: Your Base Model is Smarter Than You Thinkcited by 89
  20. 2024VisionTS: Visual Masked Autoencoders Are Free-Lunch Zero-Shot Time Series Forecasterscited by 88
  21. 2024RouterDC: Query-Based Router by Dual Contrastive Learning for Assembling Large Language Modelscited by 88
  22. 2025Time-MQA: Time Series Multi-Task Question Answering with Context Enhancementcited by 82
  23. 2025Revisiting Reinforcement Learning for LLM Reasoning from A Cross-Domain Perspectivecited by 76
  24. 2023Everything of Thoughts: Defying the Law of Penrose Triangle for Thought Generationcited by 74
  25. 2025Router-R1: Teaching LLMs Multi-Round Routing and Aggregation via Reinforcement Learningcited by 72
  26. 2025Time-VLM: Exploring Multimodal Vision-Language Models for Augmented Time Series Forecastingcited by 63
  27. 2025TimeXL: Explainable Multi-modal Time Series Prediction with LLM-in-the-Loopcited by 39
  28. 2024TSI-Bench: Benchmarking Time Series Imputationcited by 33
  29. 2024Flow of Reasoning: Training LLMs for Divergent Reasoning with Minimal Examplescited by 30
  30. 2025ArcMemo: Abstract Reasoning Composition with Lifelong LLM Memorycited by 24
  31. 2025Physics-Guided Foundation Model for Scientific Discovery: An Application to Aquatic Sciencecited by 12
  32. 2026GraphPlanner: Graph Memory-Augmented Agentic Routing for Multi-Agent LLMscited by 12
  33. 2025Time2Lang: Bridging Time-Series Foundation Models and Large Language Models for Health Sensing Beyond Promptingcited by 7
  34. 2026TimeOmni-VL: Unified Models for Time Series Understanding and Generationcited by 7
  35. 2026Stop When Reasoning Converges: Semantic-Preserving Early Exit for Reasoning Modelscited by 7
  36. 2024Time Series Analysis for Education: Methods, Applications, and Future Directionscited by 5
  37. 2025MLLM4TS: Leveraging Vision and Multimodal Language Models for General Time-Series Analysiscited by 5
  38. 2025HalluRNN: Mitigating Hallucinations via Recurrent Cross-Layer Reasoning in Large Vision-Language Modelscited by 4

50 most cited of 51 All 51 (JSON)

Built on by

Cited by year (2026 to 14 Oct)

Citing papers per year: 2026: 2.

2 papers.

  1. 2026LLMRouter: Unified Infrastructure for Developing, Evaluating, and Deploying LLM Routerscited by 4
  2. 2026TimeEvo: Failure-Driven Self-Evolution of a Time Series Agentcited by 0

Similar, not cited

Close in wording, no citation either way

  1. 2026LLaTiSA: Towards Difficulty-Stratified Time Series Reasoning from Visual Perception to Semantics
  2. 2026From Recognition to Understanding: Unlocking Cognitive Time Series Reasoning with LLMs
  3. 2025TS-Agent: Understanding and Reasoning Over Raw Time Series via Iterative Insight Gathering
  4. 2024Towards Time Series Reasoning with LLMs
  5. 2025TempoGPT: Enhancing Time Series Reasoning via Quantizing Embedding
  6. 2025TimeSense:Making Large Language Models Proficient in Time-Series Analysis
  7. 2025TimeOmni-1: Incentivizing Complex Reasoning with Time Series in Large Language Models
  8. 2026Thoth: Mid-Training Bridges LLMs to Time Series Understanding