Mystery Game Puzzles
100 positions from a puzzle version of a well-known game that Epoch AI deliberately does not name, so that no model can be prepared for it. Each position, given as text, has a single best move. Epoch AI runs it.
Hard, novel problems: graduate science, abstract puzzles, expert exams. One of 8 comparable tests in reasoning, a category the Index averages.
Read with care. Epoch AI publishes neither the game, the prompt, example positions nor model transcripts, so a result cannot be checked from outside.
Publisher: Epoch AI (internal runs)What a model is asked to do
Find the single best move in a position from a well-known game whose identity the publisher keeps secret.
For exampleThe position is described in plain text, along with whose turn it is. Exactly one legal move keeps the win. Submit it in the game's standard notation.
Why it matters. Because the game is not named, no model can have been tuned for the test. The result reflects reasoning rather than preparation.
The example is original and illustrative, not an item from the dataset.
- Models scored here
- 72
- Who produced the numbers
- Run by Epoch AI
- Items graded
- n = 100 (1.0% each)
- Best published result
- 90.0%
- Newest release covered
- November 9, 2026 (active)
- Chance level
- 9%
- License
- CC BY 4.0 (via Epoch AI)
- Direction
- Higher is better
- Scale
- Published as a share of items solved. For the Index, each score is measured as a share of the best published result (90.0%) above chance: random guessing (9%) reads 0 and the frontier reads 100, so a high-guessing exam is not flattered.
- Provenance
- Epoch AI ran the evaluation itself, under settings it documents per model, and publishes the log of every answer.
Snapshot November 10, 2026
Ranking on this test
72 models on Mystery Game Puzzles
Best published run per model, and how close each one comes to the best published result on this test (the frontier reads 100, measured above chance). Every number links to the publisher that produced it. Internal runs are marked and carry their run details on the model page. One item on this test is 1.0 points, so read gaps smaller than that as noise.
The best models, ranked here, working inside Theo.
29 of the models on this page run inside Theo today. Theo picks the right one for each step and always shows which engine answered.