VeriSoftBench (Curated Context): leaderboard
Metric: Verification success rate (%), curated context restricted to each proof's ground-truth dependency closure (an oracle-like diagnostic), on the 500 Lean 4 proof obligations, the model generating 8 candidate proof scripts per task with up to 3 repair rounds on Lean compiler feedback (specialized provers with their own internal feedback loop run end to end); higher is better. Source: arxiv.org. Saturation forecast: Around March 2027. 4 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 3 Pro | 41 | #77 |
| 2 | Claude Opus 4.5 | 31.2 | #79 |
| 3 | GPT-5.2 | 12.6 | #105 |
No result here: #3 Claude Opus 5.5, #5 GPT-6 Astra, #8 Claude Fable 5.1.
Interactive version: theaggregate.ai/benchmark?slug=verisoftbench-curated-context · How It Works · Data refreshed daily, snapshot 2026-10-11.