Claude Opus 5 with Max Reasoning Tops LMSYS Arena Leaderboards
A new model variant referred to as "Claude Opus 5" with "Max reasoning" has secured the number one spot in both the Frontend Code Arena and the Text Arena (with factuality enabled) on the LMSYS Chatbot Arena. Additionally, the "default reasoning high" configuration of the model ranked third in frontend coding and second in the text category. This achievement highlights the rapid advancement of reasoning-focused LLMs and demonstrates how integrating advanced reasoning capabilities can significantly boost performance in complex tasks like coding and factual text generation. It also showcases the competitive landscape among top AI labs, with models like Claude and Moonshot AI's Kimi K3 vying for dominance. The "factuality on" ranking metric combines human preference with factual accuracy by auditing battles, extracting verifiable claims, and verifying correctness head-to-head. While the name "Claude Opus 5" is used in the announcement, it remains unclear whether this refers to an upcoming Claude 3.5 Opus release or a typo for another version.
## BACKGROUND
The LMSYS Chatbot Arena is a widely recognized benchmarking platform that evaluates large language models (LLMs) through anonymous, crowdsourced side-by-side battles, using Elo ratings to rank them. Recently, the platform introduced specialized arenas, such as the Frontend Code Arena and a factuality-focused ranking system, to better assess models on specific, high-value capabilities.