What Does it Really Cost to Use AI for Legal Work?

Questions abound about the real cost of using artificial intelligence on legal tasks (and for work in all industries for that matter). It is no secret that AI use gets expensive and companies are spending millions, if not billions on AI token usage. As a result, at least two big name law firms recently announced they are bringing AI infrastructure in house rather than rely on other legal AI options. Some believe that at least one reason law firms are doing this is because AI is moving to consumption based pricing.

One of the datapoints we collect on Certera.ai during pairwise AI comparisons of legal LLM output is what prompts actually cost and overall token spend for legal work. You can see these numbers in real time on the Certera Legal AI Leaderboard.

So, using Certera, we decided to figure out what it really costs to run a legal question or legal workflow though an AI model. We ran 2000+ legal outputs from most available LLMs to compare their costs and get a better understanding of true legal AI pricing. 

To run the test, we used three legal prompts of varying complexity. We tested a “simple” legal query requiring only a short, direct answer, a moderately difficult task that was a little more labor intensive and required tool use (i.e. like creating a document), and finally, we tested a “hard” legal prompt requiring deep thinking and a multi-step legal workflow.

For the easy task, we used a case law question. For the “medium” task, we prompted the models to review a court opinion and draft a law firm client alert. Finally, for the most difficult task we used the coverage task from the legal evaluation dataset we released earlier this year which requires review of an insurance policy, case law and the drafting of a coverage opinion.

The high level takeaways:

• The simple legal question cost pennies. At most, it cost around $.10 USD for outputs on some OpenAI models and a fraction of a cent on open-weight models.
• The moderate task averaged $0.04 across 200 responses with a wide spread from $0.27 (Fable 5) down to fractions of a cent (Mimo).
• The difficult task was much more expensive coming in at $1.43  and costing around $10.47 a response on Fable 5.1.

You can find the full legal token and AI cost study on the Certera blog which has a detailed breakdown of costs by legal task and LLM.