The context · March 2026
OpenAI introduced GPT-5.4 mini and nano in March 2026, expanding the range of model sizes available for different tasks. [1]
A proposed project brief inspired by the period’s developments. This is an idea for exploration, not an announced Sansa product, an approved roadmap or a claim of completed client work.
Define the routine case
Choose a document type with a stable set of required fields. Use a compact model for extraction into a defined structure, then validate the result against the document’s expected format. Keep missing fields and contradictory values visible.
Design the exception route
Send ambiguous or incomplete cases to a more capable review path, with a person making the final decision when the evidence remains unclear. Include documents that look routine but contain unusual layouts. Record why each escalation happened and whether it was useful.
Compare total effort
The proposed evaluation would measure accepted extractions, missed exceptions, review time and total processing cost. Compare the two-tier workflow with a single-model baseline. Retain the routing layer only if the operational benefit justifies its extra complexity.
Source & context
OpenAI · GPT-5.4 mini and nano, 17 March 2026This retrospective was written for the archive in September 2026. The linked primary source documents the announcement or event; the practical interpretation and proposed approach are Sansa’s editorial perspective. Public examples do not imply a client relationship. Product capabilities and guidance may have changed since the period discussed.
Another perspective · March 2026
The smallest useful model is a workload question
Continue readingWorking through a similar question?
Talk it through with Sansa