AIMS: leaderboard
We argue that safety classifiers should model user intent as an explicit signal between the prompt and the final label. To study this, we introduce AIMS, a human-annotated dataset of 1,724 difficult safety prompts, each paired with an intent description and harm label. We use AIMS to evaluate intent-aware training across supervised fine-tuning, preference learning, reasoning distillation, and rein
Metric: Average (self-reported). Source: benchmarklist.com. Saturation forecast: Around July 2027. 18 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5.4 | 81.5 | #79 |
| 2 | GPT-OSS-120B | 80.3 | #341 |
| 3 | Gemma 3 12B | 80.2 | #671 |
| 4 | Claude Sonnet 4.6 | 78.2 | #90 |
| 5 | Llama 3.1 8B | 74.9 | #1151 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=aims · How It Works · Data refreshed daily, snapshot 2026-10-20.