Large language models (LLMs) are getting smarter, but are they getting safer? A groundbreaking new AI evaluation platform, Libra-Leaderboard, is shaking up the AI world by putting safety on equal footing with performance. Traditionally, AI leaderboards have focused on how well models perform tasks like writing or coding. But what about the potential for these models to spread misinformation, generate harmful content, or be vulnerable to manipulation? Libra-Leaderboard addresses this critical gap by evaluating 26 leading LLMs from organizations like OpenAI, Google, and Anthropic, across a comprehensive safety benchmark of 57 datasets. These tests cover a wide spectrum of safety risks, including bias, toxicity, information leaks, and susceptibility to adversarial attacks. The results are eye-opening, revealing significant safety vulnerabilities even in some of the most advanced models. Instead of simply averaging performance and safety scores, Libra-Leaderboard uses a unique scoring system that prioritizes balance. This encourages developers to focus on holistic improvement rather than excelling in one area at the expense of another. The platform also includes an interactive “Safety Arena” where users can test LLMs with challenging prompts and provide feedback, making AI safety accessible to a broader audience. Libra-Leaderboard isn't just about ranking models; it's about promoting responsible AI development. By highlighting the importance of safety and providing a dynamic evaluation platform, it's pushing the AI community to build safer and more trustworthy models for the future. This is a crucial step towards ensuring that the powerful potential of AI is harnessed for good, not harm.