Databricks' internal evaluations show Gemini 3.6 Flash classified documents more accurately than Gemini 3.5 Flash, particularly for long-form patents and text-heavy documents. The model achieved similar quality to high-effort parsing with approximately 32% fewer total tokens and performed well on the OfficeQA Pro agent harness benchmark.
Google's latest Flash model offers improved long-document processing and token efficiency, reducing inference costs for document-heavy AI workloads.