With advances in science and technology, optimization problems increasingly involve numerous decision variables, posing significant challenges for optimization algorithms. These problems are often mentioned as LSGO problems (large-scale global optimization) and have received considerable attention. LSGO problem is very challenging to solve due to the high dimensions, countless local optimal solutions, and huge computational burden. In this paper, a gradient approximation and hybrid grouping method (GAHG) is proposed to address such challenges. Firstly, we design an adaptive gradient approximation based search algorithm to make a quick search in a reduced search region and find preliminary good solutions. During this stage, important information about the decision variables will be stored for future usage. Further more, we introduce a hybrid grouping strategy designed to efficiently and reasonably decompose the large-scale fully non-separable problems. The hybrid grouping method consists of five decomposing strategies to assign potentially highly related variables into the same group. The performance of GAHG is evaluated on widely used large-scale benchmark suite CEC’2010 and CEC’2013 large-scale test problems. Experimental results and statistical analysis are made, which indicate the proposed GAHG is effective and efficient in solving large-scale problems.