Legora reviewed 41 documents in minutes with GPT-6 Astra

Legora, 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, reports that GPT-6 Astra completed a financial-statement tie-out across 41 documents in a single Agent run in minutes. The workflow checks figures in draft accounts against trial balances, a consolidation schedule, and the prior year’s accounts until each item agrees.

On Legora’s Benchmark for Agentic Reasoning (BAR), GPT-6 Astra improved performance by nearly 40% over the previous model on this financial-statement workflow, while averaging about a 3% improvement across all BAR tasks. In the tie-out, Astra found all four planted errors, including a £500,000 gap hidden in the revenue note, retained every correct check from the prior model, and completed around 50 more checks.

Legora stresses that the Agent handles exhaustive comparison while legal professionals retain final judgment—an approach it is extending into audit, tax, compliance, and risk.

OpenAI’s customer story is really about workflow ownership. The impressive part is not “AI reads documents”; it is that an agent can ingest a multi-document financial context, surface breaks, and leave a granular audit trail for a human reviewer. That is the difference between a demo and something a law firm might put near a filing deadline.

Signal Desk’s read: the nearly 40% gain on this specific workflow versus a ~3% average across BAR is the tell. Astra’s value here looks concentrated in long-context, high-precision comparison tasks—not a uniform leap across every legal job. Expect vendors to over-generalize from the headline number; the careful read is that financial tie-out is an especially good fit for agentic models that can keep every line item and supporting schedule in play.

What is being under-stated is liability. Keeping “final judgment with legal professionals” is necessary marketing and necessary risk allocation. If the Agent misses a break that was not planted for the demo, the expert still owns the opinion. The commercial winners will be the platforms that make the review record so clear that human oversight is fast rather than ceremonial.

Financial-statement tie-out is classic professional drudgery: Legora Legal Engineer Percevale Perks describes work that “can take an entire evening, sometimes days.” Legora’s legal engineers embed with customers to adapt the platform from contract review and legal research into adjacent professional services workflows.

This piece sits alongside other OpenAI Astra customer stories (including Playco) that argue the model’s gains show up most clearly in end-to-end agent runs rather than single-prompt benchmarks.