# ProofBench

As of Nov 3, 2026, ProofBench (v1.1) is maintained by Vals AI. Metric accuracy (%). Ceiling 100. Saturation unknown. Contamination risk low.

Declared cadence: as Vals updates (checked daily). Catalog snapshots: 1, latest Nov 3, 2026.

## Categories

- Math: display, weight 0

## Top 5

- Claude Sonnet 5.5: 100
- Claude Opus 5.5: 100
- Claude Fable 5.1: 100
- GPT-6.1 Sol: 99
- GPT-6 Astra: 99

Source: https://www.vals.ai/benchmarks/proof_bench