Methods, apparatus, and computer-readable media (transitory and non-transitory) are provided herein for reducing latency caused by switching input modalities. In various implementations, a first input such as text input may be received at a first modality of a multimodal interface provided by an electronic device. In response to determination that the first input satisfies one or more criteria, the electronic device may preemptively establish a session between the electronic device and a query processor configured to process input received at a second modality (e.g., voice input) of the multimodal interface. In various implementations, the electronic device may receive a second input (e.g., voice input) at the second modality of the multimodal interface, initiate processing of at least a portion of the second input at the query processor within the session, and build a complete query based on output from the query processor.