CHARM - Boundary Compliance: leaderboard
CHARM boundary compliance: having recognised a knowledge boundary, whether the model stays in character rather than answering from its own parametric knowledge. Accuracy, higher is better. Source: arxiv.org/abs/2609.01352.
Metric: Accuracy (%). Source: arxiv.org. Saturation forecast: Around 2031. 6 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 3.5 Flash | 64.4 | #66 |
| 2 | Qwen 3 8B | 59.5 | #678 |
| 3 | GPT-5.5 | 45.2 | #33 |
| 4 | Gemma 3 12B | 41.1 | #674 |
| 5 | Llama 3.1 8B | 37.5 | #1130 |
| 6 | GPT-4o | 18.9 | #357 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=charm-boundary-compliance · How It Works · Data refreshed daily, snapshot 2026-11-05.