Deciding between potential actions involves coordinated activity across various cortical areas such as prefrontal, premotor, parietal, and primary motor cortex as well as interconnected subcortical regions within basal ganglia (BG). In investigating the specific function of each region in this process, the role of BG remains a matter of vigorous debate. While many studies have shown decision-related activity in BG nuclei, this activity appears significantly later than in associated cortical regions, and may not be the primary driver of decisions. Also, while BG stimulation can bias choices, inactivating its main output, globus pallidus internus (GPi), does not disrupt decision-making but instead slows down movements. It is thus unclear whether BG play a causal role in decisions or instead simply control the vigor of executed movements. One possible explanation for these apparently contradictory findings is that decisions emerge within a recurrent cortico-BG circuit, where all regions contribute to the process of selecting a choice. Thus, inhibiting individual components may not lead to deficits, but stimulating any part can introduce behavioral biases. To test this hypothesis, we developed a simplified neural network model that governs the competition between potential actions while managing the trade-off between decision speed and accuracy by adjusting a context-dependent “urgency signal” in BG. Our model replicates, at both behavioral and neural levels, several key phenomena observed during recent experiments designed to dissociate the processes of deliberation and commitment. This includes decision time distributions and their variations with changes in the speed-accuracy trade-off policy, and neural activity during deliberation in prefrontal, premotor, and motor cortex, as well as GPi. The model also reproduces the effects of delayed decisions with non-specific electrical microstimulation in cortical regions, biased decisions with choice-specific optogenetic stimulation in GPi, and the limited effect of GPi inactivation on decisions made in conditions with strong sensory evidence.