Claude 3.7 Sonnet: Game-Changing AI for Legal Discovery

By John Tredennick
Two days ago, Anthropic released Claude 3.7 Sonnet, their most intelligent model to date and the market’s first hybrid reasoning model. Our Chief Data Scientist and team have been working around the clock to test this breakthrough technology, and the results are remarkable.
After extensive testing, we can confidently state that Claude 3.7 Sonnet is the most powerful model available for legal work, particularly for summarizing documents such as deposition transcripts, medical records or finding, analyzing and reporting on large document sets through advanced GenAI-integrated systems like our DiscoveryPartner® platform.
Real-World Testing with TREC Topics
We added Claude Sonnet 3.7 to our DiscoveryPartner platform this morning (after two days of testing in our labs). I couldn’t resist taking Claude out for a spin so I ran a test using one of the TREC Topics employed by NIST, analyzing a large collection of Jeb Bush emails.
Here was the prompt I used, based on the original topic created by Gordon Cormack for the annual TREC program:
We want to know about this topic: Condominiums — All documents concerning the rules and organizations governing Florida condominium associations and conflicts between owners and managers in Florida. Relevant documents include those concerning the establishment of the Florida office of ombudsman, and issues relating to hiring and firing the ombudsman.
○ Who was the ombudsman?
○ Why was he hired?
○ Why was he fired?
○ Why was he reinstated?
○ Who were the key witnesses who knew about this topic?
——-
I started with a population of about 275,000 documents, asking DiscoveryPartner to find the top 300 most likely relevant documents (using three kinds of search: semantic, keyword algorithmic and a CAL classifier). I then instructed Claude Haiku 3.5 to read and summarize the documents before passing them on to its big brother, Claude Sonnet 3.5 for analysis.
Why use Haiku for summaries? Simple. We have found that Haiku does a great job of summarizing individual documents but at a cost about 30 times less than its bigger sibling. So using multiple LLM models for different tasks is smarter and far more cost effective. That’s why we went multi-LLM from the beginning.
In less than two minutes, DiscoveryPartner identified the 300 most relevant documents, summarized them, and leveraged Claude 3.7 Sonnet to produce a comprehensive analysis. The results blew me away (and I’m a hardened GenAI user).
Here’s an excerpt from the response:
Florida Condominium Ombudsman: Establishment, Leadership, and Controversies
Establishment of the Florida Condominium Ombudsman Office
The Florida Condominium Ombudsman position was created by the Florida Legislature in 2004, sponsored by Representative Julio Robaina and Senator Walter “Skip” Campbell ( Bush1283689¶4, Bush719129¶11-¶12, Bush908166¶4). This was the first national condo ombudsman position ( Bush719129¶9) and was established to address escalating conflicts in Florida’s growing condominium market ( Bush1389566¶5-6, Bush1464386¶5, Bush1464284¶2).
The office was created despite objections from condo lawyers and directors ( Bush719129¶11) and was intended to create a “level playing field” for owners and boards ( Bush719129¶15-¶16). The position was established in response to the perceived ineffectiveness of the Department of Business and Professional Regulation (DBPR) in addressing condominium issues ( Bush1349171¶4, Bush1471927¶4, Bush1582020¶4).
The ombudsman position was funded by a $4 per unit annual collection from condo associations ( Bush719129¶20, Bush679444¶10-12, Bush1474868¶7). Initially, the legislature did not provide specific spending authority for the Ombudsman’s office ( Bush861228¶3, Bush863180¶14). By 2005, the legislature provided $444,029 and six positions for the Ombudsman’s office ( Bush895505¶7, Bush895512¶8).
Dr. Virgil Rizzo: Florida’s First Condominium Ombudsman Appointment
Dr. Virgil Rizzo, a 67-year-old retired doctor and lawyer, was appointed as Florida’s first condominium ombudsman by Governor Jeb Bush on December 6, 2004 ( Bush719129¶8-¶10, Bush861228¶2, Bush718833¶3, Bush863392¶20, Bush866355¶3).
Rizzo was born in Pittsburgh in 1937, educated as both a medical doctor and lawyer, served in the military during the Vietnam War, and had been a condominium resident since 1971 ( Bush718833¶9-¶12). He had connections to Democratic state Senator Walter “Skip” Campbell, who sponsored the legislation creating the position ( Bush861065¶25-¶26).
Jan Bergemann advocated for Rizzo’s appointment, describing him as a condo owner who had personally experienced association problems, not affiliated with any trade organizations dealing with associations, and independent and capable of giving an “honest try” to the position ( Bush710166¶3).
Role and Responsibilities
Rizzo’s primary mission was to mediate disputes between condo owners and associations and reduce “condo commando” conflicts ( Bush861065¶7). The ombudsman was responsible for educating over 1 million condo owners and resolving disputes with 17,000 condo associations ( Bush719129¶9).
Rizzo established three departments: dispute resolution between boards and owners, owner and director education, and recommending new laws and procedures ( Bush719129¶13). The ombudsman was required to coordinate educational materials for condo owners ( Bush895512¶9) and monitor condominium board elections when requested ( Bush957713¶5, Bush1217779¶4).
By 2006, the office handled about 700 complaints and requests weekly ( Bush1389566¶9) and had grown to six staff members with a $417,000 budget ( Bush1472104¶16). . . .
The report continues for a total of 8 single-spaced pages. You can see the entire unedited report here. The whole process from finding the top 300 relevant documents, to summarizing and synthesizing the information, took two minutes. “We aren’t in Kansas anymore, Toto.”
Why This Matters
You know why this matters. I just did a week’s worth of research and analysis on a topic I know absolutely nothing about in minutes. The cost for that work went from $10,000+ to $200. The game is changing and the impact of technology like this will be felt high and low across the legal industry and beyond.
To be fair, It’s important to note that this level of document analysis cannot be achieved with Claude 3.7 Sonnet alone. This test showcases the unique capabilities of DiscoveryPartner to:
- Rapidly identify the most relevant documents from massive collections
- Generate meaningful summaries that capture essential information
- Leverage Claude 3.7 Sonnet’s reasoning to synthesize findings into actionable intelligence
At the same time, the report is written by Claude and I share it just to give you an idea of the power of a GenAI system like Claude for analyzing large document sets.
The Future Is Now
Generative AI technology continues to improve at a breathtaking pace, with significant advancements emerging almost daily. The combination of DiscoveryPartner with Claude 3.7 Sonnet represents a fundamental shift in how legal professionals can find, analyze and report on information in large document sets.
For law firms and legal departments seeking to maintain a competitive edge, embracing these technologies isn’t just about efficiency—it’s about delivering superior results for clients through enhanced analytical capabilities.
Would you like to see how DiscoveryPartner and Claude 3.7 Sonnet can transform your legal discovery process? Contact us for a demonstration.

About the Author
John Tredennick (jt@merlin.tech) is the CEO and Founder of Merlin Search Technologies, a software company leveraging generative AI and cloud technologies to make investigation and discovery workflow faster, easier, and less expensive. Prior to founding Merlin, Tredennick had a distinguished career as a trial lawyer and litigation partner at a national law firm.
With his expertise in legal technology, he founded Catalyst in 2000, an international ediscovery technology company that was acquired in 2019 by a large public company. Tredennick regularly speaks and writes on legal technology and AI topics, and has authored eight books and dozens of articles. He has also served as Chair of the ABA’s Law Practice Management Section.