Optimizing the performance of non-premixed burners to simultaneously achieve high thermal efficiency, low pollutant emissions, and maintain robust operational stability requires computational fluid dynamics (CFD) predictive tools that accurately model turbulence-chemistry interactions (TCI) at the microscale. Conventional Reynolds-Averaging Navier-Stokes (RANS) models often suffer from poor capture of microscale effects for predicting thermal NOx formation and near-quenching phenomena. This study, employing a dual numerical approach, couples an adapted RANS solver with a large eddy simulation (LES) framework equipped with a flame-generated manifold model (FGM) and the GRI-Mech 3.0 kinetic mechanism, and validates both against experimental data. This combined approach is used to predict the performance of a fuel-air coflow burner under systematic variations in the air-to-fuel mass flow ratio (MFR). Time-frequency analysis of fluid and thermodynamic oscillations identified a dominant low-frequency (10–50 Hz) hydrodynamic instability related to the collapse of vortices in the flow separation zone. Quantitative results showed that while co-current airflow-maintained carbon conversion efficiency (CCE) above 80% up to MFR=5 and strongly reduced NOx emissions through thermal dilution, increasing the dilution ratio resulted in reduced thermal stability and increased risk of sustained hydrodynamic oscillations. This comparative modeling approach provides a robust predictive framework for the design of safer and more efficient industrial burners. The LES-FGM model demonstrated superior accuracy in capturing TCI, flame stability boundaries, and thermal NOx sensitivity compared to RANS, especially in high-strain regions.