miniF2F

What it measures

Formalized mathematical statements test whether a system can construct proofs accepted by a proof assistant.

How to read the result

A proof checker supplies a sharper validity boundary than a free-form assessment of a natural-language proof.

Limits and comparison traps

Pin prover version, library, search budget, and statement formulation. Formal validity proves the formal statement, which may still differ from an intended informal claim.

Published evaluation

Pass@1 over the 244-statement test set. Formal systems remain part of the model identity; validation scores and pass@8 are separate metrics.

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

Pass@1 over the 244-statement test set. Formal systems remain part of the model identity; validation scores and pass@8 are separate metrics.

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.