The asymmetry of language switch cost has been the focus in the filed of language switch. Previous studies found that the switching cost is symmetrical in proficient bilinguals, but asymmetrical in less proficient bilinguals. Therefore, some researchers augured that language switch cost is affected by L2 proficiency. However, others proposed that inhibitory control is the real cause of language switch cost. In short, many models have been proposed to explain this linguistic phenomenon from different perspectives. Inhibitory control model and language-specific selection model are two classical theories. The former holds that lexical selection in the two languages are realized through the practice of inhibition, the degree of inhibition leads to symmetrical or asymmetrical language switch cost. On the contrary, the latter insists that language switch cost is modulated by L2 proficiency. In addition, other models are attempting to explain the two classical theories from different angles. Task-set inertia model is consistent with inhibitory control model, both of which claim that inhibitory control plays a key role in language switch. Furthermore, sequential difficulty effect model emphasizes language proficiency, which is in line with language-specific selection model. Therefore, no agreement has been reached yet on the asymmetry of language switch cost. Cognitive neuroscience studies on inhibitory control and L2 proficiency have revealed the neural mechanisms of language switch. The fMRI studies on less proficient bilinguals showed that language switch activated the right prefrontal cortex (rPFC), which reflects the role of inhibitory control, indicating that less proficient bilinguals need to rely on inhibitory mechanism to complete language switch successfully, and supporting the inhibitory control model. Nevertheless, some studies on proficient bilinguals found that rPFC was not activated, so they believed that language proficiency may modulate the symmetry of language switch cost, and that language-specific selection model was correct. Moreover, cognitive neuroscience studies found that prefrontal cortex, anterior cingulate cortex (ACC) and left caudate nucleus were activated in language switch and task switch, indicating that the cost of language switch and task switch might be identical to some extent. Future research directions on language switch were pointed out in this paper. Specifically, (1) How do inhibition and language proficiency affect language switch? If the language switch cost was symmetrical for high inhibitory control of less proficient bilinguals and proficient bilinguals, inhibitory control will be the real reason for language switch cost. If the language switch cost in proficient bilinguals was symmetrical, and asymmetrical for high inhibitory control of non-proficient bilinguals, L2 proficiency will be an important factor for language switch cost. (2) Are language switch cost and task switch cost the same nature? If the patterns of proficient and less proficient bilinguals in language switch cost and task switch cost are parallel, then the nature of the two tasks will be the same. If the language switch cost in proficient and less proficient bilinguals differs from task switch cost, language switch cost will at least be specific. (3) Can the inhibitory mechanism of domain-general task switch be generalized to language switch? If language switch cost in non-proficient bilinguals was symmetrical after training, domain-general inhibitory training could be generalized to language switch. Otherwise, L2 proficiency will be the key factor to language switch cost.