Part 7/16:
- Refusal Bench: 57.1%, significantly outpacing GPT-4 (17.67%) and Claude 4 (17%), demonstrating a more balanced and neutral stance on delicate topics.
This last point reflects Noose Research's philosophy: their models are designed to handle sensitive and controversial prompts responsibly, engaging thoughtfully rather than outright refusing or censoring.