Open math problem solving is critical to help students deepen the understanding and promote transfer of mathematics knowledge. However, the cognitive mechanism for open math problem solving, particularly the role of spatial abilities, has not been paid enough attention. This study recruited 192 junior middle school students (14.30 ± 0.48 years old). Results showed that both spatial visualisation (measured by paper folding) and spatial working memory (measured by spatial 2-back) significantly associated with open math problem solving with controlled variables including age, sex, nonverbal matrix reasoning, and arithmetic principles. Moreover, spatial working memory was more associated with easy open math problem solving, while spatial visualisation more associated with difficult open math problem solving. These findings suggested that as the difficulty of open math problems increased, the construction of problem space mattered more than retaining and updating problem space in open math problem solving.