Epoch AI - Proofbench (v1.0): leaderboard
The same ProofBench problems under the superseded v1.0 grading. Epoch's externally-run copy keeps the whole history in one file and marks each carried-over model 'v1.0-era value; model absent from the v1.1 board', because vals.ai's August 2026 re-grade changed how proofs are marked and upstream states the two sets of numbers are not comparable. Split out so a v1.0 score is never ranked beside a v1.1 one; verifier-checked proof success rate (%).
Metric: Score. Source: epoch.ai. 35 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | claude-opus-4-8_max | 69 | |
| 2 | gpt-5.4-2026-03-05_xhigh | 56 | |
| 3 | claude-opus-4-7_max | 54 | |
| 4 | gpt-5.5_xhigh | 50 | |
| 5 | claude-opus-4-6_max | 50 | |
| 6 | claude-sonnet-4-6_max | 45 | |
| 7 | muse-spark-1.1 (xHigh) | 39 | |
| 8 | claude-opus-4-5-20251101_unknown (High) | 36 | |
| 9 | gemini-3.6-flash_unknown (High) | 36 | |
| 10 | glm-5.2_max | 35 |
Interactive version: theaggregate.ai/benchmark?slug=epoch-ai-proofbench-v1-0 · How It Works · Data refreshed daily, snapshot 2026-10-20.