In a customer case study published September 3, 2026, OpenAI reported that legal-tech firm Legora used GPT-6 Astra to complete a financial-statement tie-out across 41 documents in a single Agent run. Legora describes itself as an agentic operating system for legal and professional work, used by more than 100,000 professionals across more than 1,800 in-house legal departments and law firms in over 50 markets. Financial-statement tie-out involves checking every figure in draft accounts against trial balances, a consolidation schedule and the previous year's accounts. Legora Legal Engineer Percevale Perks said the work "can take an entire evening, sometimes days." According to the account, the Agent checked every balance against its supporting schedule, surfaced breaks in the amounts and recorded each check, with the human expert responsible for the final judgment on each result. Legora evaluated the model using its Legora Benchmark for Agentic Reasoning (BAR). It reports GPT-6 Astra improved performance by nearly 40% over the previous model on this financial-statement workflow, while the improvement averaged about 3% across all BAR tasks. Legora also says the model found all four errors it had planted in the accounts, including a £500,000 gap hidden in the revenue note, and completed around 50 more checks than the previous model. The figures come from OpenAI and Legora; no independent verification was cited.
- Tie-out across 41 documents completed in a single Agent run, reported as minutes
- Legora reports nearly 40% improvement on this workflow vs. the previous model; ~3% average across all BAR tasks
- Model found all four planted errors, including a £500,000 gap in the revenue note; human retains final judgment
What it means for you
This is a vendor showing off: OpenAI and Legora are demonstrating that a new model can speed up a tedious accounting-review task that normally takes an evening or more. The numbers come from the companies themselves, not an outside test, so treat them as marketing rather than proof. The one durable idea worth keeping is the pattern — let the machine do the exhaustive line-by-line checking, keep a human making the final call.
Try this
If your business does any repetitive reconciliation — invoices against statements, figures against a schedule — try running one small batch through an AI tool and have a person verify every flagged item. Note how many real errors it catches versus false alarms before trusting it further.
Who should care
Accountants, bookkeepers and finance staff at small firms who spend hours cross-checking numbers, plus legal and audit teams already using Legora.
Skip this if
You don't do document-heavy reconciliation work, or you're not a Legora customer — this is a case study for a specific platform, not a general tool you can pick up today.
Sources: OpenAI — read the original