HoS Benign Utility - GSM8K 1-shot: leaderboard
Heads of Security compares ordinary question answering and reasoning after training models to resist prompt injection. Accuracy (%).
Metric: Accuracy (%). Source: arxiv.org. Saturation forecast: Estimated already saturated. 7 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Llama 3.3 70B Instruct | 93.7 | #533 |
| 2 | GPT-OSS-20B | 93.6 | #517 |
No result here: #3 Claude Opus 5.5, #5 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=hos-benign-utility-gsm8k-1-shot · How It Works · Data refreshed daily, snapshot 2026-10-29.