Published on September 10, 2026, the account describes a document-purpose classifier trained on 200 examples labeled by an agent. The model classified 191,724 documents for about US$0.70 in inference compute.
In an account published on September 10, 2026, the author describes a workflow combining an agent, SetFit, and Jobs to build a document-purpose classifier. The agent labeled 200 examples used in the process.
The model classified 191,724 FinePDFs-Edu documents for about US$0.70 in inference compute. The example shows how a small classifier can reduce the cost of large-scale data curation. To check the method, scope, and results, consult the original publication and verify the experiment details there; the available summary does not report accuracy metrics.