Financial-statement tie-out is the kind of work that makes junior associates question their career choices. Checking every figure in draft accounts against trial balances, consolidation schedules, and prior-year statements is tedious, error-prone, and — as Legora Legal Engineer Percevale Perks puts it — can “take an entire evening, sometimes days.” Legora just ran that workflow through GPT-6 Astra, and the results are hard to ignore.
In a single Agent run, Legora’s system reviewed 41 documents, checked every balance against its supporting schedule, and surfaced breaks in the amounts. It found all four errors Legora had deliberately planted, including a £500,000 gap hidden in the revenue note. Perks attributes the jump to raw processing power: “the ability to ingest such a large number of documents, digest really complex information, and get all of those different line items and figures.” Against Legora’s own benchmark for that workflow, performance improved by nearly 40% over the previous model.
What’s more telling is what didn’t break. The Agent retained every check the prior model got right and completed roughly 50 additional checks. That’s the reliability story nobody writes press releases about — not just new capabilities, but not regressing on what already worked. Across all tasks in Legora’s Benchmark for Agentic Reasoning, the average improvement was a more modest 3%, which suggests the tie-out workflow is a sweet spot for Astra’s long-context reasoning rather than evidence of a uniform leap everywhere.
The human stays in the loop by design. Legora’s pitch — and it’s the right one for legal tech — is that the Agent handles exhaustive comparison while the expert owns the judgment call on each result. Legora sees this as the template for pushing beyond law into audit, tax, compliance, and risk. With 100,000-plus professionals already using the platform across 1,800 legal departments and firms, that expansion isn’t theoretical. The question is whether a 3% average gain across all tasks justifies the compute cost, or whether Astra’s value concentrates in a handful of document-heavy workflows where 40% improvements actually change how people bill their hours.
