Glossary · Term

GPQA

← all terms

Definition

Plain language

A small set of extremely hard graduate-level science questions used to test AI reasoning.

As stated in the literature

GPQA-Diamond, the hardest subset of the Graduate-Level Google-Proof Q&A benchmark, covering expert-validated physics, chemistry, and biology questions.

Also called: GPQA-Diamond

Why it matters: It tests whether models can handle expert-level science questions that aren't easily Googleable, separating real reasoning from search-and-paraphrase.

For example, a GPQA question might ask the experimental technique needed to distinguish two reaction mechanisms in organometallic chemistry — answerable by an expert but resistant to web search.

Heard on the show

“On GPQA Diamond, a graduate-level science exam, the reasoning model wins by about nineteen points.”
Episode 197 — Twin Problems Suggest AI Reasoning Gains Are Mostly Better Fact Recall

Mentioned in 5 episodes

  1. 197
    Twin Problems Suggest AI Reasoning Gains Are Mostly Better Fact Recall
  2. 140
    When a Reasoning Model Says "Let Me Double-Check" After It's Already Decided
  3. 094
    Chain-of-Thought Monitoring Fails Across Languages, and Worst Where It's Needed Most
  4. 077
    Reading a Model's Confidence Curve to Decide When Chain-of-Thought Is Worth It
  5. 058
    Why Upgrading Your AI Auditor to a Smarter Model Can Make Your System Less Safe

Related concepts

Related terms