Google Research's latest blog post describes a decomposition strategy for intent extraction. Instead of asking a model to classify intent in one step, the task is broken into smaller, more manageable subtasks. According to the post, this approach yields superior results while using small models.
The central claim is that model size is not the only path to better performance. By structuring the problem carefully, compact models can match or exceed the results of larger systems. The post frames this as a practical direction for efficient AI.
Since this is a single source, there are no conflicting reports to compare. The post is a research announcement rather than an independent evaluation, so the results should be read as the authors' findings.