Bonmu Ku, a graduate student at Harvard, knows what a perfect LSAT score demands. He earned one himself. Now, his research shows that artificial intelligence can do the same.
Ku tested eight AI models on every officially licensed LSAT exam available, covering more than a decade of real administrations. Four of the eight achieved a perfect scaled score of 180. OpenAI’s GPT-5, Anthropic’s Claude Opus 4, DeepSeek-R1, and Moonshot’s Kimi K2 Thinking all reached the maximum. It is the first time any AI system has hit the ceiling of the exam under independently verified conditions.
The LSAT is widely regarded as one of the most difficult standardized tests in existence. Roughly 170,000 candidates sit for it each year, and it remains the most influential factor in law school admissions. Fewer than 0.1% of test-takers ever score a perfect 180.
The leap is recent. In March 2023, OpenAI reported that GPT-4 had scored a 163, placing it at the 88th percentile. Three years later, four models have reached the top. The distance between the 88th percentile and a perfect score may appear small on paper, but on the LSAT’s compressed upper scale, it reflects a dramatic jump in reasoning consistency.
What changed is the architecture. The models that reached 180 are part of a new generation of reasoning systems. Instead of predicting the next most likely word, they produce an internal chain of thought, working through each step before arriving at an answer. Ku’s ablation experiments confirmed that when this thinking capability was disabled, performance fell significantly. On a test that rewards sustained logical precision, that gap proved decisive.
“The LSAT doesn’t test what you know. It tests how you think,” Ku said. “There’s no facts to memorize, no case law to recall. It isolates pure logical reasoning, the kind of thinking that law schools believe only the sharpest legal minds possess. AI now has it.”
Beyond the headline results, Ku experimented with a technique called process reward modeling, which grades a model’s reasoning at every step along the way, not just at the end. The method produced measurable gains in scores, raising the possibility that even the models that fell short of 180 could eventually get there.
The timing is notable. As AI reshapes the job market and displaces white-collar workers, more people are turning to law school as a stable career path. Applications are surging, with the 2025–2026 cycle seeing a 33% increase in submissions and a 22% rise in test registrations. The very technology driving candidates toward the LSAT is now outperforming them on it.
The LSAT was built to identify the sharpest legal minds in the country. That benchmark, however, no longer belongs to humans alone. Ku would know. He has been on both sides of a perfect score.
