I found a more accurate upcoming LLM Benchmarks leaderboard (www.llmverifications.com)

A newly announced leaderboard for large language models (LLMs) showcases upcoming benchmark results, offering detailed metrics on various LLMs’ performance. Claude Opus 5.5 by Anthropic tops the list with an impressive accuracy score of 85.0%, successfully answering 28 out of 30 science questions and excelling in multiple categories, including engineering and humanities. Following closely are Google's Gemini 4 Argon and OpenAI's GPT-5.3-Codex, reflecting a high level of competition among leading models. This leaderboard is significant for the AI and machine learning community as it provides transparent and independent evaluations of LLMs, focusing on aspects such as accuracy, speed, latency, and cost-effectiveness. The unbiased comparisons allow developers and researchers to make informed decisions when selecting models for their specific applications. The benchmarks reveal not only the technical capabilities of these models but also serve as a catalyst for ongoing improvements in AI systems, fostering innovation in model design and deployment.