OpenTSLM TeeMoE: A Unified Time-Series Language Model for Forecasting, Contextual Prediction, and Reasoning

Real-world time-series applications increasingly require models to move between and often combine numerical forecasting, context-conditioned prediction, and language-based temporal reasoning. Yet current time-series foundation models remain fragmented across these capabilities: numerical specialists often provide the strongest forecasts, while language-based models offer broader contextual understanding and analysis. A central challenge is to unify these heterogeneous capabilities without reducing their individual performance. We introduce OpenTSLM TeeMoE, a generalist time-series language model that independently trains low-rank experts for forecast aggregation, native forecasting, and temporal analysis over a shared backbone, rather than learning all capabilities through one monolithic adapter. A learned controller then dynamically composes the frozen expert updates according to each request. Our unified model can forecast directly from observed time series, reason over textual context and temporal patterns, and synthesize and refine predictions from external numerical forecasting specialists. At the time of submission, the composed model ranks second on GIFT-Eval (mean MASE rank 19.2062), third on Context is Key (RCRPS 0.11607), and first on TimeSeriesExam (78.69% accuracy).