Pearl Finds AI Models Match Expert Judgment Only 70% of the Time

Leading AI models from OpenAI, Anthropic, Google DeepMind, Microsoft, and other developers align with expert judgment only about 70% of the time, according to a press release from Pearl Enterprise. The company evaluated 25 models using more than 500 professional questions across five domains: business, health, law, pets, and technology.

Pearl’s Expert Alignment Leaderboard found that OpenAI’s GPT 5.5 led with 72.7% expert alignment, followed closely by GPT 5 at 72.5%, GPT 5.1 at 72.0%, and Anthropic’s Claude Opus 4.7 at 71.9%. No model exceeded 73% overall alignment, indicating that current systems may be converging below expert-level performance.

Performance varied sharply by domain. Top scores reached 80.9% in business, but some widely used models dropped to around 20% in areas such as law and health. Pearl also noted that increasing reasoning depth improved results by only up to 2.6 percentage points, and in some cases reduced response quality.

Each model received identical prompts and was scored on correctness, completeness, prioritization, and professional judgment. Pearl stated that the dataset used for evaluation was not previously available to model developers. The full leaderboard is available on Pearl’s website.