Cortega team · November 2026
TL;DR
The model bill for legal research is small. Claude Opus 5.5 answered 100 legal research questions for $4.88 in our test. The cost that matters is the lawyer time needed to find and fix the wrong citations in those answers.
This article puts numbers on that cost and compares three ways to handle it.
Take a law firm that verifies 50,000 citations a year across the documents it produces. In our question set, 100 answers contained about 1,590 citations, so 50,000 citations is about 31 batches of 100 answers.
The firm uses Claude Opus 5.5 as its agent. Without a check, nobody knows which citations are wrong. In our test, 11.8% of Opus's citations were wrong, and 58 of 100 answers contained at least one wrong citation.
A lawyer at $150 an hour has two realistic ways to deal with this.
A1. Verify by hand. The lawyer opens every citation, confirms the case exists and says what the answer claims, asks the model to rewrite the failing answers, and checks the rewrites. At 2 minutes per citation, this costs about $12,570 per 100 answers.
A2. Use a checker tool and prompt the model. A checker flags the bad citations. The lawyer writes the rewrite prompt, reads the new answer, and runs the checker again. We assumed 2 to 3 rounds for an answer that needs a rewrite, 8 minutes of lawyer time per round, and 10 minutes of final review per answer. This costs about $5,415 per 100 answers.
Both versions depend on the lawyer catching every error. We have not measured how many wrong citations remain after this process.
With Cortega, the check, rewrite and re-check run automatically. No person writes a prompt or reads a report in between. An open-weight model does the checking, which keeps the check cheap.
| Setup | Cost per 100 answers | Wrong citations after the loop |
|---|---|---|
| Claude Sonnet 5.5 + Cortega AI Verifier | $11.74 | 0.2% |
| Open-weight model + Cortega AI Verifier | $5.34 | 0.8% |
For the 50,000-citation firm, about 31 batches of 100 answers:
| Approach | Approximate annual cost |
|---|---|
| A1. Lawyer verifies by hand | $396,000 |
| A2. Checker tool and manual rewrites | $170,000 |
| Open-weight model + Cortega AI Verifier | $170 |
The Cortega figure covers the model, the check and the rewrites. Lawyer time spent finding and fixing errors is the real cost of AI in legal work.
Without a check, the best single model in our test still had 7.6% of citations wrong. A firm with a strict accuracy bar cannot reach it by choosing a better model alone. With Cortega, both setups above reached under 1%.
If your bar is looser, around 8% wrong, switching to a stronger model alone is enough, and you do not need Cortega for that.
MAKE YOUR NEXT MOVE
Explore how Cortega fits your AI environment.