OpenAI case study: GPT-6 Astra clears a 41-document financial review in minutes and catches every planted error

OpenAI published a case study on September 3 showing how Legora, an agentic operating system for legal and professional work, used GPT-6 Astra to complete a labor-intensive financial-review task in minutes.

Legora says it is used by more than 100,000 professionals across over 1,800 in-house legal departments and law firms in more than 50 markets, and its legal engineers adapt the platform to customer workflows ranging from contract review to legal research.

The case centers on financial-statement tie-out: checking every figure in draft accounts against trial balances, a consolidation schedule, and the previous year's accounts until each item agrees, work that Legora legal engineer Percevale Perks says can take an entire evening and sometimes days.

Using GPT-6 Astra, Legora's Agent completed the tie-out across 41 documents in a single run and within minutes, checking every balance against its supporting schedule, surfacing breaks in the amounts, and recording each check so the legal professional is left with a granular record of every line item and figure.

In an evaluation tied to Legora's Benchmark for Agentic Reasoning, which measures end-to-end legal tasks drawn from real-world use cases, GPT-6 Astra improved performance by nearly 40% over the previous model on this financial-statement workflow, while the average gain across all BAR tasks was about 3%.

The model found all four errors Legora had planted in the accounts, including a £500,000 gap hidden in the revenue note, and it retained every check the previous model got right while completing around 50 additional checks.

The message for the industry is that agentic review can now absorb entire document sets in a single pass, with the human expert left to make the final call.

What to watch next: whether this kind of tie-out result holds up across real audit cycles, and how rivals respond with benchmark claims of their own.