Legora reviewed 41 documents in minutes with GPT-6 Astra

Legora, an agentic operating system for legal and professional work used by over 100,000 professionals, recently tested OpenAI's GPT-6 Astra on a tedious financial-statement tie-out workflow. According to Legora, financial tie-outs traditionally require an entire evening or even days to check figures in draft accounts against supporting schedules. Utilizing GPT-6 Astra, the Legora Agent completed the review across 41 documents in a single run within minutes. Legora evaluated the model using its Benchmark for Agentic Reasoning, reporting that GPT-6 Astra improved performance by nearly 40% on this specific financial workflow, while average improvements across all benchmark tasks were about 3%. In testing, the Agent successfully surfaced breaks, checked every balance against its supporting schedule, and recorded each check. Notably, the Agent found all four errors planted by Legora, which included a £500,000 gap hidden in the revenue note. The company notes that the system retains checks from previous models while completing roughly 50 additional ones, delivering a faster and more complete first pass. Legora emphasizes that while the Agent handles the exhaustive comparison, human experts remain responsible for the final judgment calls. This human-in-the-loop methodology supports Legora's expansion into audit, tax, compliance, and risk workflows.