Wake5 - Vulnerability Recall - Tools On: leaderboard
Wake5 v1 Table5 static-analysis tool ablation on 134 paired cases: recall 55 positive cases, correct rejection 79 negative cases. Native printed percentage extracted from its retained numerator/denominator cell. Tools on/off are distinct configurations. Sol/Terra/Luna are the paper’s short endpoint labels, not a guessed API generation. Precision with response-dependent denominator, cost, speed and agent-written fuzzer coverage are withheld.
Metric: Accuracy / success (%). Source: arxiv.org. Saturation forecast: Rough model projection: around 2026. 3 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Sol (Wake5 Run: static-analysis=Tools On;endpoint=source-short-label;effort=unreported) | 69.1 | |
| 2 | Terra (Wake5 Run: static-analysis=Tools On;endpoint=source-short-label;effort=unreported) | 58.2 | |
| 3 | Luna (Wake5 Run: static-analysis=Tools On;endpoint=source-short-label;effort=unreported) | 58.2 |
Interactive version: theaggregate.ai/benchmark?slug=wake5-vulnerability-recall-tools-on · How It Works · Data refreshed daily, snapshot 2026-11-12.