40 original questions — abstract matrices, formal logic, number series, verbal analogies — administered through our own unified LLM API at temperature 0. Every answer, right or wrong, is published below.
First run in progress · 40 items staged · temp 0 · methodology
Leaderboard pending — first evaluation run in progress.
Every model’s score, per-item choices, and response excerpts publish here after the run. The question bank and protocol are already public.
Scores compare frontier LLMs — OpenAI and Anthropic models — on identical items in a single run. They are not human IQ norms, and they say nothing about API latency, token pricing, or coding benchmarks — see how the test works.
Raven-style 3×3 pattern matrices serialized to text — rotation, counts, fills, XOR and distribution rules.
10 items · difficulty 1-5
Deduction: syllogisms, modus tollens, knights and knaves, seating and assignment puzzles.
10 items · difficulty 1-5
Number series with layered rules: arithmetic, geometric, interleaved and recursive progressions.
10 items · difficulty 1-5
Analogies, antonym and synonym pairs, odd-one-out, and collective nouns.
10 items · difficulty 1-5
The questions, prompts, and raw responses are all public. Generate a key, rerun the same benchmark on the same models — or on your own — and compare numbers. $5 in free credit is enough for several full runs.
A linear mapping of difficulty-weighted accuracy onto a 70-145 scale: 0% correct maps to 70, 100% correct maps to 145. It is not a human-normed IQ score — it exists to compare models against each other under identical questions.
The first run is in progress. The leaderboard covers the frontier models sold on this platform — OpenAI and Anthropic tiers — and every score, per-item choice, and response excerpt publishes here once the run completes.
Yes. The full question bank, the exact prompts, and every model response are published. Generate an API key, point the harness at the OpenAI-compatible endpoint, and you will get the same items in the same order at temperature 0.