DPrivBench - Algorithm Level: leaderboard
Metric: F1 score (0-1, times 100) of the yes verdict on the DPrivBench Category 2 algorithm-level instances (125 algorithms from the differential privacy literature with perturbed negative versions); binary yes/no verdict on whether the stated guarantee holds, five seeds, no tools; higher is better. Source: arxiv.org. Saturation forecast: Around January 2027. 14 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | GPT-5.5 (High) | 82.9 |
| 2 | Gemini 3.1 Pro (Preview) | 81 |
| 3 | GPT-5.4 (High) | 79.3 |
| 4 | Gemini 3 Pro | 74.8 |
| 5 | GPT-5 (High) | 74.2 |
| 6 | DeepSeek R1 | 61.4 |
| 7 | DeepSeek V3.1 (Non-reasoning) | 60.3 |
| 8 | Gemini 2.5 Flash | 60.2 |
| 9 | GPT-5 (Minimal) | 54.8 |
| 10 | Claude Sonnet 4.5 | 51.8 |
| 11 | Claude Opus 4.5 | 48.7 |
| 12 | Qwen 3 30B A3B 2507 Instruct | 46 |
| 13 | Qwen 3 30B A3B 2507 (Thinking) | 40.7 |
Interactive version: theaggregate.ai/benchmark?slug=dprivbench-algorithm-level · How It Works · Data refreshed daily, snapshot 2026-10-07.