What it measures

10-problem public inspection release of FrontierMath with full Inspect log traces.

Reading the measurement

A correct final answer, a complete proof and useful research progress are different achievements. Choose the one the task actually grades. Inspect the permitted computation, attempt budget and handling of invalid arguments before comparing models.

Published evaluation

Source-reported per-run best score across scorers, converted from fractions to percentages. This is the ten-question public sample, not the 290-question private set used for headline FrontierMath scores. Task-version strings remain visible. Epoch later corrected public problem PLD1 without retroactively updating the public-sample scores, so these historical values are not a corrected current leaderboard.

Where this fits in an evaluation suite

Final-answer accuracy, proof credit, and expert-reviewed research progress are distinct signals. A proof-aware evaluation should retain partial work and audit strong claims.

Start with a capability hypothesis, then choose a second test with a different failure mode. Keep the checkpoint, reasoning effort, agent tools, sampling policy, and scoring procedure fixed when measuring a training intervention. Record failures and missing outputs in the denominator.

A reproducible comparison

Source-reported per-run best score across scorers, converted from fractions to percentages. This is the ten-question public sample, not the 290-question private set used for headline FrontierMath scores. Task-version strings remain visible. Epoch later corrected public problem PLD1 without retroactively updating the public-sample scores, so these historical values are not a corrected current leaderboard.

Publish the exact run configuration, item coverage, score units, and available uncertainty. If a source supplies only an aggregate or a single run, preserve that limit. A new model name or a rounded score alone does not establish a capability improvement.

PRIMARY EVIDENCE & FURTHER READING

  1. A Rosetta Stone for AI Benchmarks — released Epoch evaluation runs ↗

    Epoch AI · 2025

Reviewed 2026-10-27. Explanation and interpretation by RLTasks. Benchmark ownership remains with its original authors.