BALTIMORE, June 29 — For families watching artificial intelligence creep into classrooms, a new study out of the University of Maryland Francis King Carey School of Law offers a surprising reality check: the latest AI might not be acing law school exams any better than last year’s model. The human side of this story is one professors and students alike are watching closely. Last year, researchers gave OpenAI’s o3 reasoning model a set of final exams, and the results turned heads.
The AI scored anywhere from an A+ to a B, graded on the same curve as the law students. It was a moment that made many wonder: is the bar exam next?
But the sequel, published this week, tells a different tale. This year, the same team repeated the experiment with OpenAI’s updated model, GPT-5.5. What they found was not a leap forward, but something close to a plateau.
The newer model did not noticeably outperform last year’s version, and in some respects, it may have done worse. “This apparent plateau, despite a better model and using extra-high inference-time compute, may reflect trends observed in other legal benchmarks for AI and makes theoretical sense,” the authors concluded.
That is a sobering line for anyone who assumed each new release would bring a straight line of improvement. What this means for patients, for families, for anyone whose legal fate might one day be touched by AI: the technology is still learning the law, and it may be hitting a wall. Behind the breakthrough — or the lack of one — is a question that goes beyond test scores.
The researchers are not stopping here. They plan to continue the experiment in future semesters, tracking whether the AI’s performance evolves or stalls. The TaxProf Blog had coverage of the findings, and the discussion among legal educators has been lively.
Some law professors, meanwhile, have already begun moving to old-school assessment methods to combat AI. The study raises a parallel question: if AI can take the exams, should it also be grading them?
The same researchers have explored whether artificial intelligence can be used for grading law school exams, but the answer remains unsettled. For now, the takeaway is a human one. The AI is not getting worse, but it is not getting much better either.
That plateau, as the researchers call it, may be a natural limit — or it may be a temporary pause before the next leap. Either way, the experiment is far from over.
The study comes from a law school, not a tech lab, and that matters. It is one thing to benchmark AI on a multiple-choice bar prep test. It is another to hand it a professor’s final exam, grade it on the same curve as students, and watch it earn a B or an A+.
That is a real-world test, and the results this year suggest the curve might be flattening. The authors did not say why the newer model failed to pull ahead. They pointed to broader trends in legal AI benchmarks and noted that the finding makes theoretical sense.
That is a careful, academic way of saying: we are not sure yet, but we are watching. And for the students sitting in those law school classrooms, that might be the most reassuring news of all.
{“@context”: “https://schema.org”, “@type”: “NewsArticle”, “headline”: “Is AI getting better at law school exams? Maybe not, new study shows”, “description”: “BALTIMORE, June 29 — For families watching artificial intelligence creep into classrooms, a new study out of the University of Maryland Francis King Carey School…”, “datePublished”: “2026-06-29 15:58:00T09:30:00+00:00”, “author”: {“@type”: “Person”, “name”: “James Roberto”}, “publisher”: {“@type”: “Organization”, “name”: “InfoPulse Today”, “url”: “https://newsfeed.infopulsetoday.com/”}}
