We present a hybrid halo model, which improves the description of the 2-halo term by incorporating non-linear information from simulations. A linear computation of the halo-halo power spectrum is inaccurate at the transition between the 1-halo and 2-halo regimes, whereas nonlinear approaches such as Hybrid Effective Field Theory (HEFT) are not naturally compatible with the halo model decomposition. We address this limitation by constructing a collapsed HEFT (CHEFT) framework, in which the power-spectrum templates of the HEFT operator expansion are measured from simulations where 1-halo contributions are removed by collapsing particles to their halo centres. The halo-halo power spectrum is then expressed as a sum over bias operators, with mass-dependent bias parameters deduced from simulation using the probabilistic bias approach. This provides a predictive model in which there are no free bias parameters. We validate the model for a range of weighting schemes designed to mimic the halo-mass dependence of astrophysical observables, including the Sunyaev-Zeldovich effect, the Cosmic Infrared Background, and galaxy abundances described via a halo occupation distribution. For the matter field, the model recovers the power spectrum to percent-level accuracy across the transition regime. For weighted tracers, the baseline model achieves accuracies of $\sim 5-10\%$ in power, which improves to the $\sim 3-5\%$ level when including an effective higher-derivative, Laplacian-like contribution in the bias expansion. The CHEFT model thus retains the precision and flexibility of the EFT approach, while allowing the transparent incorporation of astrophysical effects that are directly associated with haloes.