OpenAI's technical report for GPT-5, published in full on Thursday, confirms what the benchmark community had been predicting for several months: the model achieves performance at or above the ninetieth percentile of human test-takers on sixteen professional licensing examinations, including the Uniform Bar Examination, the Certified Public Accountant examination, the United States Medical Licensing Examination, and the Chartered Financial Analyst Level III examination.

The results arrive at a moment of particular sensitivity for the professional services sector. Bar associations in several jurisdictions have spent the past eighteen months debating the appropriate scope of AI-assisted legal practice; the CPA profession has been grappling with the implications for audit methodology; and medical licensing boards have convened emergency working groups on the question of AI in clinical decision support.

What the benchmark figures do not resolve — and what the professional bodies are acutely aware of — is the distinction between passing an examination and competently practising a profession. The examination assesses a particular kind of structured knowledge retrieval. The profession requires judgment under uncertainty, contextual sensitivity, ethical navigation, and accountability.