Forecasting in biological systems presents different considerations and difficulties from traditional time-series settings, most notably the high-dimensionality associated with modern biological assays. A comprehensive analysis of the performance of modern forecasting methods on biological datasets is currently missing from the literature, in particular one that evaluates potential gains from larger model sizes and their increased complexity. Here, we assess 14 models spanning 4 complexity scales (Baseline, Statistical, Neural, and LLM-based) on 5 time-series datasets. We show that model scale and complexity does not uniformly improve performance across biological datasets, and that in some cases, highly complex models fail to outperform common baselines.