OpenAI reports that Legora, a financial-review platform, used GPT-6 Astra to work through 41 documents in a matter of minutes. The exercise centered on a statement-review workflow, where the model was expected to spot problems in the material.
The model found all four deliberately planted errors and delivered a nearly 40% improvement in performance over the prior process. That points to a concrete benefit for document-heavy financial tasks: faster review without missing known issues.
Because the only source is OpenAI's own write-up, the results are vendor-reported rather than independently verified. Still, the demonstration offers a clear example of how large multimodal models might be applied to professional document review in finance.