AI Is Closing In on Expert-Level Knowledge Work — Faster Than Anyone Expected
OpenAI's GDPval benchmark is quietly redefining how we measure AI against human professionals — and the results should have every knowledge worker paying attention.
Released in September 2025 and significantly updated in December, GDPval tests frontier AI models on real, economically valuable tasks drawn from 44 professions across high-GDP industries. We're talking financial modeling, legal drafting, marketing strategy, engineering reports, presentations, spreadsheets — the kind of multi-hour deliverables that companies pay senior professionals handsomely to produce.
These aren't sanitized academic exercises. Tasks were designed by industry experts averaging roughly 14 years of experience, and evaluation was rigorous: human professionals, working blind, compared AI-generated outputs against actual expert deliverables and rated them as better, equal, or worse.
The Results Are Hard to Ignore
Early frontier models were already competitive. Claude Opus 4.1 led the initial wave with a ~47.6% win-or-tie rate against human experts — meaning it matched or outperformed seasoned professionals in nearly half of all tasks.
Then came GPT-5.2 in December 2025, which crossed a critical threshold: a 70.9% win-or-tie rate against top industry professionals. That makes it the first model to perform at or above expert level across GDPval's full task set.
The economics are just as striking. AI-generated outputs were produced more than 11x faster and at less than 1% of the cost of a human expert. Add light human oversight — a review-and-refine workflow rather than creation from scratch — and the productivity multiplier becomes enormous.
What This Actually Means
This isn't about chatbots getting marginally better at answering questions. It's about AI systems handling substantive, high-value professional work — the kind that currently fills the calendars of analysts, consultants, lawyers, and engineers.
The progress curve so far has been steep and remarkably linear. If that trajectory holds, the implications for teams, roles, and entire industries over the next 12–24 months are profound.
The question isn't whether AI will reshape knowledge work. It's how quickly your organization adapts.