A new study introduces a diffusion-based generative approach for joint bias correction and downscaling of subseasonal temperature forecasts, addressing the coarse spatial resolution and systematic biases that limit their direct use in decision-making for sectors such as energy planning, agriculture, and public health. Subseasonal forecasting is defined as predicting atmospheric and surface variables from 2 weeks to 2 months ahead, filling a gap between medium-range weather prediction and seasonal forecasts. Unlike conventional post-processing methods such as quantile mapping, which correct forecasts based on marginal distributions and ignore case-specific forecast–observation relationships, the proposed method learns a conditional distribution of high-resolution temperature fields given the coarse-resolution forecast and lead time, enabling case-dependent probabilistic post-processing. The framework is demonstrated over Switzerland, where strong elevation gradients and local climate effects pose substantial challenges for forecasting systems. The study finds that while traditional statistical methods remain competitive in probabilistic skill, this work establishes diffusion-based generative modeling as a viable foundation for next-generation subseasonal post-processing. By identifying current capabilities and remaining limitations, the study outlines clear directions for future development toward high-resolution subseasonal predictions that can better support decision-making.