GABRIEL: OpenAI Turns Quantitative Encoding of Unstructured Materials into an Auditable Research Tool
OpenAI Economic Research released the GABRIEL paper and open-source tool in February 2026, converting qualitative attributes in text, images, and audio into repeatable quantitative measurements, with batch processing, retries, checkpoints, and audit trails.
LLM-assisted research has long been stuck with one-off annotation scripts. GABRIEL organizes prompts, batching, recovery, validation, and logging into a standard tool, enabling social science and industry research to treat model judgments as a measurement process that requires calibration and auditing.
The tool covers rating, ranking, classification, extraction, deduplication, and de-identification, and the paper validates GPT annotation quality across multiple task types. The official repository continuously releases versions, indicating the research has transitioned from methodology to reusable software assets.
Consulting, market research, policy analysis, and knowledge management teams can process large-scale qualitative materials at lower cost, but must retain sampling review, model versioning, source evidence, and bias assessment.
When using LLMs for batch research, audit trails, checkpoint recovery, and human calibration capabilities like those in GABRIEL should be included in procurement requirements to avoid non-reproducible one-off analyses.
Key areas to watch: reliability in cross-language and high-risk domains, measurement drift across different model versions, and whether third parties can reproduce the accuracy and bias conclusions in the paper.