This paper contributes to growing debates over the decision-making ability of democracy by considering the epistemic value of deliberative democracy. It focuses on the benefits democratic deliberation can derive from its diversity, and the extent to which these benefits can be realised with respect to the complexities of political problems. The paper first calls attention to the issue of complexity through a critique of Hélène Landemore and the Diversity Trumps Ability Theorem. This approach underestimates complexity due to its reliance on an ‘oracle assumption’ and this is shown to highlight more general difficulties for applying the benefits of diversity to the realities of political problems. The paper then develops a new model of deliberation—based on an relationship between cognitive diversity and diminishing returns to cognitive type—which does not involve an oracle assumption and can therefore support the epistemic value of deliberative democracy even for complex problems. The benefits of diversity are also argued to be better realised though sortition than either democratic elections or epistocracy, pointing to the value of deliberation between randomly selected citizens. Finally, and contrary to past work, the new account suggests that diversity cannot alone establish the superiority of democratic deliberation over all non-democratic alternatives, and that it is insufficient to mount a purely epistemic argument for deliberative democracy. The paper therefore furthers our understanding of the epistemic value of deliberative democracy by clarifying when and to what extent diversity is a benefit to political problem solving.